fix(asr): normalize origin rejection statuses

cai

2026-08-11 db6463d9f98f740d4e8b6d3a0a25b65688196321

 src/main.rs

@@ -4,6 +4,7 @@

use std::{
    borrow::Cow,
    collections::HashSet,
    env, fs,
    path::{Path, PathBuf},
    sync::{
@@ -14,7 +15,9 @@
};

use anyhow::{Context, Result, anyhow};
use asr_realtime::{RealtimeAsrConfig, RealtimeAsrOutcome, RealtimeAsrUpload};
use asr_realtime::{
    AudioIngressMetadata, RealtimeAsrConfig, RealtimeAsrOutcome, RealtimeAsrUpload,
};
use audio::{AudioDiagnostics, load_pre_recorded_frames};
use base64::{Engine as _, engine::general_purpose};
use futures_util::StreamExt;
@@ -27,12 +30,13 @@
    options::TrackPublishOptions,
    prelude::{
        DataPacket, LocalAudioTrack, LocalTrack, ParticipantIdentity, RemoteAudioTrack,
        RemoteTrack, Room, RoomEvent, RoomOptions,
        RemoteParticipant, RemoteTrack, Room, RoomEvent, RoomOptions,
    },
};
use reqwest::Client;
use serde::{Deserialize, Serialize};
use serde_json::json;
use sha2::{Digest, Sha256};
use tokio::time::{sleep, sleep_until, timeout};
use tokio::{
    sync::{mpsc, mpsc::UnboundedReceiver, watch},
@@ -50,6 +54,11 @@
const PCM_16K_SAMPLE_RATE_HZ: u32 = 16_000;
const BOT_NUM_CHANNELS: u16 = 1;
const TRACK_NAME: &str = "bot-main-audio";
const STREAM_TIMING_VERSION: u32 = 1;
const STREAM_TIMING_FIRST_SEGMENT: u64 = 1;
const STREAM_TIMING_FIRST_CHUNK: u64 = 1;
const STREAM_TIMING_MAX_ELAPSED_MS: u64 = 5_000;
const STREAM_TIMING_SOURCE: &str = "stream_anchor_monotonic";

#[tokio::main(flavor = "multi_thread")]
async fn main() -> Result<()> {
@@ -663,6 +672,31 @@
    println!("{payload}");
}

fn emit_anchored_activity(
    call_id: &str,
    trace_id: &str,
    turn_id: &str,
    event_name: &str,
    marker: &RuntimeTurnStreamTimingMarker,
    extension: serde_json::Value,
) {
    let payload = json!({
        "type": "cv_activity",
        "callId": call_id,
        "traceId": trace_id,
        "turnId": turn_id,
        "eventName": event_name,
        "eventWallTimeMs": current_time_millis(),
        "serverDeltaMs": marker.server_delta_ms,
        "serverDeltaSource": STREAM_TIMING_SOURCE,
        "result": "ok",
        "reasonCode": null,
        "retryable": null,
        "extension": marker.extension_with(extension),
    });
    println!("{payload}");
}

fn spawn_user_audio_observer(
    events: UnboundedReceiver<RoomEvent>,
    config: &Config,
@@ -675,6 +709,7 @@
    let enabled = config.user_audio_observer_enabled;
    let simple_vad_enabled = config.simple_vad_enabled;
    let simple_vad_config = config.simple_vad_config.clone();
    let user_participant_identity = config.user_participant_identity.clone();
    let turn_bridge_config = TurnBridgeConfig::from_config(config);

    tokio::spawn(async move {
@@ -696,9 +731,14 @@
            turn_bridge_config,
            http,
            sink,
            user_participant_identity,
        )
        .await;
    })
}

fn is_bound_user_participant(identity: &str, expected: Option<&str>) -> bool {
    expected.is_none_or(|value| identity == value)
}

async fn observe_user_audio_events(
@@ -711,6 +751,7 @@
    turn_bridge_config: TurnBridgeConfig,
    http: Client,
    sink: Arc<BotAudioOutputSink>,
    user_participant_identity: Option<String>,
) {
    info!(
        call_id = %call_id,
@@ -734,6 +775,15 @@
                publication: _,
                participant,
            } => {
                if !is_bound_user_participant(
                    &participant.identity().to_string(),
                    user_participant_identity.as_deref(),
                ) {
                    warn!(call_id = %call_id, trace_id = %trace_id,
                        metadata_status = "wrong_participant",
                        "runtime helper ignored non-user audio participant");
                    continue;
                }
                let participant_alias = redact(&participant.identity().to_string());
                let track_sid_alias = redact(&track.sid().to_string());
                let track_name = track.name();
@@ -759,6 +809,8 @@
                    turn_bridge_config.clone(),
                    http.clone(),
                    sink.clone(),
                    user_participant_identity.clone(),
                    participant,
                );
            }
            RoomEvent::TrackSubscribed {
@@ -953,7 +1005,11 @@
            .await
            {
                Ok(outcome) => {
                    let mut published_device_outputs = HashSet::new();
                    for output in &outcome.device_outputs {
                        if !should_publish_device_output(&mut published_device_outputs, output) {
                            continue;
                        }
                        if let Err(error) = sink
                            .publish_device_output(call_id, trace_id, &turn.turn_id, output)
                            .await
@@ -1419,6 +1475,7 @@
            .await?;
        }
    }
    state.close_timing();
    if !state.completed {
        warn!(
            call_id = %call_id,
@@ -1548,6 +1605,7 @@
                call_id,
                trace_id,
                turn,
                &event,
                audio_chunk,
                state,
                turn_pipeline_started_at,
@@ -1559,6 +1617,9 @@
        }
        Some("device_output") => {
            if let Some(output) = event.device_output.as_ref() {
                if !should_publish_device_output(&mut state.published_device_output_ids, output) {
                    return Ok(());
                }
                sink.publish_device_output(call_id, trace_id, &turn.turn_id, output)
                    .await?;
                state.device_output_count = state.device_output_count.saturating_add(1);
@@ -1566,6 +1627,7 @@
        }
        Some("turn_completed") => {
            state.completed = true;
            state.close_timing();
            info!(
                call_id = %call_id,
                trace_id = %trace_id,
@@ -1576,6 +1638,7 @@
            );
        }
        Some("turn_failed") => {
            state.close_timing();
            let error = event.error.as_ref();
            let reason_code = error
                .and_then(|value| value.reason_code.as_deref())
@@ -1610,6 +1673,7 @@
        }
        Some("turn_cancelled") => {
            state.completed = true;
            state.close_timing();
            info!(
                call_id = %call_id,
                trace_id = %trace_id,
@@ -1642,6 +1706,19 @@
        }
        Some("activity") => {
            if let Some(activity) = event.activity.as_ref() {
                if activity.event_type.as_deref() == Some("tts_first_audio_chunk_ready") {
                    if let Some(runtime_session_nonce) =
                        bridge_config.runtime_session_nonce.as_deref()
                    {
                        let _ = state.arm_timing_anchor(
                            call_id,
                            trace_id,
                            &turn.turn_id,
                            runtime_session_nonce,
                            &event,
                        );
                    }
                }
                info!(
                    call_id = %call_id,
                    trace_id = %trace_id,
@@ -1700,6 +1777,7 @@
    call_id: &str,
    trace_id: &str,
    turn: &FinishedSpeechTurn,
    event: &RuntimeTurnStreamEvent,
    audio_chunk: &RuntimeTurnStreamAudioChunk,
    state: &mut RuntimeTurnStreamState,
    turn_pipeline_started_at: Instant,
@@ -1719,6 +1797,58 @@
        .unwrap_or("pcm_s16le")
        .trim()
        .to_ascii_lowercase();
    if !matches!(format.as_str(), "pcm_s16le" | "mp3" | "mpeg" | "wav") {
        return Err(anyhow!("unsupported reply_audio_chunk format {format}"));
    }
    match state.reply_chunk_markers.observe(audio_chunk.segment_seq) {
        ReplyChunkMarker::FirstReply => {
            let extension = json!({
                "segmentSeq": audio_chunk.segment_seq,
                "chunkSeq": audio_chunk.chunk_seq,
                "format": format.as_str(),
                "bytes": payload.len(),
            });
            if let Some(marker) =
                state.record_m6(call_id, trace_id, &turn.turn_id, event, audio_chunk)
            {
                emit_anchored_activity(
                    call_id,
                    trace_id,
                    &turn.turn_id,
                    "helper_first_reply_audio_chunk_received",
                    &marker,
                    extension,
                );
            } else {
                emit_activity(
                    call_id,
                    trace_id,
                    Some(&turn.turn_id),
                    "helper_first_reply_audio_chunk_received",
                    "ok",
                    None,
                    None,
                    extension,
                );
            }
        }
        ReplyChunkMarker::SegmentFirst => emit_activity(
            call_id,
            trace_id,
            Some(&turn.turn_id),
            "helper_segment_first_audio_chunk_received",
            "ok",
            None,
            None,
            json!({
                "segmentSeq": audio_chunk.segment_seq,
                "chunkSeq": audio_chunk.chunk_seq,
                "format": format.as_str(),
                "bytes": payload.len(),
            }),
        ),
        ReplyChunkMarker::None => {}
    }
    let frames = if format == "pcm_s16le" {
        let sample_rate = audio_chunk.sample_rate.unwrap_or(sink.sample_rate_hz);
        let channels = audio_chunk.channels.unwrap_or(u32::from(sink.num_channels));
@@ -1810,7 +1940,7 @@
            Err(error) => return Err(error).context("failed to decode final stream audio chunk"),
        }
    } else {
        return Err(anyhow!("unsupported reply_audio_chunk format {format}"));
        unreachable!("supported encoded format checked above")
    };
    if frames.is_empty() {
        return Ok(0);
@@ -1838,21 +1968,33 @@
        sink.write_pcm_frame(frame).await?;
        if !state.first_audio_frame_written {
            state.first_audio_frame_written = true;
            emit_activity(
                call_id,
                trace_id,
                Some(&turn.turn_id),
                "bot_reply_first_audio_frame_written",
                "ok",
                None,
                None,
                json!({
                    "replyPlaybackMode": state.reply_playback_mode.as_str(),
                    "format": format.as_str(),
                    "chunkSeq": audio_chunk.chunk_seq,
                    "replyTotalAfterVadEndMs": turn_pipeline_started_at.elapsed().as_millis() as u64,
                }),
            );
            let extension = json!({
                "replyPlaybackMode": state.reply_playback_mode.as_str(),
                "format": format.as_str(),
                "chunkSeq": audio_chunk.chunk_seq,
                "replyTotalAfterVadEndMs": turn_pipeline_started_at.elapsed().as_millis() as u64,
            });
            if let Some(marker) = state.record_m7() {
                emit_anchored_activity(
                    call_id,
                    trace_id,
                    &turn.turn_id,
                    "bot_reply_first_audio_frame_written",
                    &marker,
                    extension,
                );
            } else {
                emit_activity(
                    call_id,
                    trace_id,
                    Some(&turn.turn_id),
                    "bot_reply_first_audio_frame_written",
                    "ok",
                    None,
                    None,
                    extension,
                );
            }
        }
        sleep_until(pacing_started_at + Duration::from_millis(((index + 1) as u64) * 20)).await;
    }
@@ -2306,9 +2448,12 @@
    completed: bool,
    audio_chunk_count: u64,
    device_output_count: u64,
    published_device_output_ids: HashSet<String>,
    encoded_audio_buffer: Vec<u8>,
    pcm_stream_decoder: Option<audio::PcmS16leStreamDecoder>,
    pcm_stream_network_chunk_count: u64,
    reply_chunk_markers: ReplyChunkMarkerState,
    timing: RuntimeTurnStreamTimingState,
}

impl Default for RuntimeTurnStreamState {
@@ -2321,10 +2466,266 @@
            completed: false,
            audio_chunk_count: 0,
            device_output_count: 0,
            published_device_output_ids: HashSet::new(),
            encoded_audio_buffer: Vec::new(),
            pcm_stream_decoder: None,
            pcm_stream_network_chunk_count: 0,
            reply_chunk_markers: ReplyChunkMarkerState::default(),
            timing: RuntimeTurnStreamTimingState::default(),
        }
    }
}

impl RuntimeTurnStreamState {
    fn arm_timing_anchor(
        &mut self,
        call_id: &str,
        trace_id: &str,
        turn_id: &str,
        runtime_session_nonce: &str,
        event: &RuntimeTurnStreamEvent,
    ) -> bool {
        if self.timing.phase != RuntimeTurnStreamTimingPhase::Empty
            || event.call_id.as_deref() != Some(call_id)
            || event.trace_id.as_deref() != Some(trace_id)
            || event.turn_id.as_deref() != Some(turn_id)
        {
            return false;
        }
        let Some(activity) = event.activity.as_ref() else {
            return false;
        };
        if activity.event_type.as_deref() != Some("tts_first_audio_chunk_ready") {
            return false;
        }
        let Some(extension) = activity.extension.as_ref() else {
            return false;
        };
        let Some(anchor_id) = extension.stream_anchor_id.as_deref() else {
            return false;
        };
        let valid_anchor_id = (16..=64).contains(&anchor_id.len()) && anchor_id.is_ascii();
        let expected_nonce_hash = runtime_session_nonce_hash(runtime_session_nonce);
        if extension.stream_timing_version != Some(STREAM_TIMING_VERSION)
            || !valid_anchor_id
            || extension.runtime_session_nonce_hash.as_deref() != Some(expected_nonce_hash.as_str())
            || extension.segment_seq != Some(STREAM_TIMING_FIRST_SEGMENT)
            || extension.stream_timing_validation.as_deref() != Some("bound")
        {
            return false;
        }
        let Some(server_delta_ms) = extension.stream_anchor_server_delta_ms else {
            return false;
        };
        self.timing.anchor = Some(RuntimeTurnStreamTimingAnchor {
            call_id: call_id.to_string(),
            trace_id: trace_id.to_string(),
            turn_id: turn_id.to_string(),
            anchor_id: anchor_id.to_string(),
            server_delta_ms,
            runtime_session_nonce_hash: expected_nonce_hash,
            segment_seq: STREAM_TIMING_FIRST_SEGMENT,
            received_at: Instant::now(),
        });
        self.timing.phase = RuntimeTurnStreamTimingPhase::Armed;
        true
    }

    fn record_m6(
        &mut self,
        call_id: &str,
        trace_id: &str,
        turn_id: &str,
        event: &RuntimeTurnStreamEvent,
        audio_chunk: &RuntimeTurnStreamAudioChunk,
    ) -> Option<RuntimeTurnStreamTimingMarker> {
        if self.timing.phase != RuntimeTurnStreamTimingPhase::Armed {
            return None;
        }
        let anchor = self.timing.anchor.as_ref()?;
        if anchor.call_id != call_id
            || anchor.trace_id != trace_id
            || anchor.turn_id != turn_id
            || event.call_id.as_deref() != Some(call_id)
            || event.trace_id.as_deref() != Some(trace_id)
            || event.turn_id.as_deref() != Some(turn_id)
            || audio_chunk.segment_seq != Some(anchor.segment_seq)
            || audio_chunk.chunk_seq != Some(STREAM_TIMING_FIRST_CHUNK)
            || audio_chunk.stream_timing_version != Some(STREAM_TIMING_VERSION)
            || audio_chunk.stream_anchor_id.as_deref() != Some(anchor.anchor_id.as_str())
        {
            return None;
        }
        let Some(marker) = RuntimeTurnStreamTimingMarker::from_anchor(
            anchor,
            audio_chunk.chunk_seq.unwrap_or(STREAM_TIMING_FIRST_CHUNK),
        ) else {
            self.close_timing();
            return None;
        };
        self.timing.chunk_seq = Some(marker.chunk_seq);
        self.timing.phase = RuntimeTurnStreamTimingPhase::M6Recorded;
        Some(marker)
    }

    fn record_m7(&mut self) -> Option<RuntimeTurnStreamTimingMarker> {
        if self.timing.phase != RuntimeTurnStreamTimingPhase::M6Recorded {
            return None;
        }
        let anchor = self.timing.anchor.as_ref()?;
        let Some(marker) = RuntimeTurnStreamTimingMarker::from_anchor(
            anchor,
            self.timing.chunk_seq.unwrap_or(STREAM_TIMING_FIRST_CHUNK),
        ) else {
            self.close_timing();
            return None;
        };
        self.timing.phase = RuntimeTurnStreamTimingPhase::M7Recorded;
        Some(marker)
    }

    fn close_timing(&mut self) {
        self.timing.close();
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum RuntimeTurnStreamTimingPhase {
    Empty,
    Armed,
    M6Recorded,
    M7Recorded,
    Closed,
}

struct RuntimeTurnStreamTimingState {
    phase: RuntimeTurnStreamTimingPhase,
    anchor: Option<RuntimeTurnStreamTimingAnchor>,
    chunk_seq: Option<u64>,
}

impl Default for RuntimeTurnStreamTimingState {
    fn default() -> Self {
        Self {
            phase: RuntimeTurnStreamTimingPhase::Empty,
            anchor: None,
            chunk_seq: None,
        }
    }
}

impl RuntimeTurnStreamTimingState {
    fn close(&mut self) {
        self.anchor = None;
        self.chunk_seq = None;
        self.phase = RuntimeTurnStreamTimingPhase::Closed;
    }
}

impl Drop for RuntimeTurnStreamTimingState {
    fn drop(&mut self) {
        self.anchor = None;
        self.chunk_seq = None;
    }
}

struct RuntimeTurnStreamTimingAnchor {
    call_id: String,
    trace_id: String,
    turn_id: String,
    anchor_id: String,
    server_delta_ms: u64,
    runtime_session_nonce_hash: String,
    segment_seq: u64,
    received_at: Instant,
}

struct RuntimeTurnStreamTimingMarker {
    version: u32,
    anchor_id: String,
    anchor_server_delta_ms: u64,
    anchor_elapsed_ms: u64,
    server_delta_ms: u64,
    runtime_session_nonce_hash: String,
    segment_seq: u64,
    chunk_seq: u64,
}

impl RuntimeTurnStreamTimingMarker {
    fn from_anchor(anchor: &RuntimeTurnStreamTimingAnchor, chunk_seq: u64) -> Option<Self> {
        let elapsed_ms = u64::try_from(anchor.received_at.elapsed().as_millis()).ok()?;
        if elapsed_ms > STREAM_TIMING_MAX_ELAPSED_MS {
            return None;
        }
        Some(Self {
            version: STREAM_TIMING_VERSION,
            anchor_id: anchor.anchor_id.clone(),
            anchor_server_delta_ms: anchor.server_delta_ms,
            anchor_elapsed_ms: elapsed_ms,
            server_delta_ms: anchor.server_delta_ms.checked_add(elapsed_ms)?,
            runtime_session_nonce_hash: anchor.runtime_session_nonce_hash.clone(),
            segment_seq: anchor.segment_seq,
            chunk_seq,
        })
    }

    fn extension_with(&self, extra: serde_json::Value) -> serde_json::Value {
        let mut extension = match extra {
            serde_json::Value::Object(value) => value,
            _ => serde_json::Map::new(),
        };
        extension.insert("streamTimingVersion".to_string(), json!(self.version));
        extension.insert("streamAnchorId".to_string(), json!(self.anchor_id));
        extension.insert(
            "streamAnchorServerDeltaMs".to_string(),
            json!(self.anchor_server_delta_ms),
        );
        extension.insert("anchorElapsedMs".to_string(), json!(self.anchor_elapsed_ms));
        extension.insert(
            "runtimeSessionNonceHash".to_string(),
            json!(self.runtime_session_nonce_hash),
        );
        extension.insert("segmentSeq".to_string(), json!(self.segment_seq));
        extension.insert("chunkSeq".to_string(), json!(self.chunk_seq));
        extension.insert("streamTimingValidation".to_string(), json!("bound"));
        serde_json::Value::Object(extension)
    }
}

fn runtime_session_nonce_hash(value: &str) -> String {
    let digest = Sha256::digest(value.as_bytes());
    digest[..6]
        .iter()
        .map(|byte| format!("{byte:02x}"))
        .collect()
}

#[derive(Debug, PartialEq, Eq)]
enum ReplyChunkMarker {
    FirstReply,
    SegmentFirst,
    None,
}

#[derive(Default)]
struct ReplyChunkMarkerState {
    first_reply_seen: bool,
    seen_segments: HashSet<u64>,
}

impl ReplyChunkMarkerState {
    fn observe(&mut self, segment_seq: Option<u64>) -> ReplyChunkMarker {
        let first_for_segment = segment_seq
            .map(|value| self.seen_segments.insert(value))
            .unwrap_or(false);
        if !self.first_reply_seen {
            self.first_reply_seen = true;
            return ReplyChunkMarker::FirstReply;
        }
        if first_for_segment {
            return ReplyChunkMarker::SegmentFirst;
        }
        ReplyChunkMarker::None
    }
}

@@ -2344,6 +2745,12 @@
struct RuntimeTurnStreamEvent {
    #[serde(rename = "type", alias = "event")]
    event_type: Option<String>,
    #[serde(rename = "callId")]
    call_id: Option<String>,
    #[serde(rename = "traceId")]
    trace_id: Option<String>,
    #[serde(rename = "turnId")]
    turn_id: Option<String>,
    #[serde(rename = "seq")]
    seq: Option<u64>,
    #[serde(rename = "replyPlaybackMode")]
@@ -2379,6 +2786,12 @@
struct RuntimeTurnStreamAudioChunk {
    #[serde(rename = "chunkSeq", alias = "seq")]
    chunk_seq: Option<u64>,
    #[serde(rename = "segmentSeq")]
    segment_seq: Option<u64>,
    #[serde(rename = "streamTimingVersion")]
    stream_timing_version: Option<u32>,
    #[serde(rename = "streamAnchorId")]
    stream_anchor_id: Option<String>,
    format: Option<String>,
    #[serde(rename = "sampleRate")]
    sample_rate: Option<u32>,
@@ -2395,6 +2808,23 @@
    stage: Option<String>,
    #[serde(rename = "reasonCode")]
    reason_code: Option<String>,
    extension: Option<RuntimeTurnStreamTimingExtension>,
}

#[derive(Clone, Deserialize)]
struct RuntimeTurnStreamTimingExtension {
    #[serde(rename = "streamTimingVersion")]
    stream_timing_version: Option<u32>,
    #[serde(rename = "streamAnchorId")]
    stream_anchor_id: Option<String>,
    #[serde(rename = "streamAnchorServerDeltaMs")]
    stream_anchor_server_delta_ms: Option<u64>,
    #[serde(rename = "runtimeSessionNonceHash")]
    runtime_session_nonce_hash: Option<String>,
    #[serde(rename = "segmentSeq")]
    segment_seq: Option<u64>,
    #[serde(rename = "streamTimingValidation")]
    stream_timing_validation: Option<String>,
}

#[derive(Deserialize)]
@@ -2427,6 +2857,30 @@
    #[serde(rename = "commandCode")]
    command_code: Option<String>,
    params: Option<serde_json::Value>,
}

fn should_publish_device_output(
    published_ids: &mut HashSet<String>,
    output: &RuntimeTurnDeviceOutput,
) -> bool {
    let Some(command_id) = output
        .command_id
        .as_deref()
        .map(str::trim)
        .filter(|value| !value.is_empty())
    else {
        return false;
    };
    if output
        .command_code
        .as_deref()
        .map(str::trim)
        .filter(|value| !value.is_empty())
        .is_none()
    {
        return false;
    }
    published_ids.insert(command_id.to_string())
}

fn require_safe_segment(value: &str) -> Result<()> {
@@ -2468,6 +2922,8 @@
    turn_bridge_config: TurnBridgeConfig,
    http: Client,
    sink: Arc<BotAudioOutputSink>,
    expected_participant_identity: Option<String>,
    participant: RemoteParticipant,
) -> JoinHandle<()> {
    tokio::spawn(async move {
        let mut stream = NativeAudioStream::new(
@@ -2546,6 +3002,7 @@
            None
        };
        let mut realtime_asr_upload: Option<RealtimeAsrUpload> = None;
        let mut last_fixture_sequence: Option<String> = None;

        while let Some(drained) = frame_rx.recv().await {
            let frame = drained.frame;
@@ -2579,8 +3036,11 @@

            if let Some(vad) = simple_vad.as_mut() {
                if vad_enabled_gate.load(Ordering::Acquire) {
                    let was_in_speech = vad.in_speech;
                    let turn = vad.observe_frame(
                    let participant_identity = participant.identity().to_string();
                    let (was_in_speech, is_in_speech, turn) = observe_bound_participant_frame(
                        &participant_identity,
                        expected_participant_identity.as_deref(),
                        vad,
                        &call_id,
                        &trace_id,
                        &participant_alias,
@@ -2588,40 +3048,15 @@
                        frame_count,
                        elapsed_ms,
                        &frame,
                        http.clone(),
                        turn_bridge_config.realtime_asr_config(),
                        || participant.attributes(),
                        &mut realtime_asr_upload,
                        &mut last_fixture_sequence,
                        turn_bridge_config.asr_realtime_enabled,
                    );
                    let is_in_speech = vad.in_speech;

                    if !was_in_speech && is_in_speech {
                        let turn_id = format!("turn-{:04}", vad.turn_index);
                        match RealtimeAsrUpload::start(
                            http.clone(),
                            turn_bridge_config.realtime_asr_config(),
                            &call_id,
                            &trace_id,
                            &turn_id,
                            &vad.speech_samples,
                        ) {
                            Ok(upload) => {
                                info!(
                                    call_id = %call_id,
                                    trace_id = %trace_id,
                                    turn_id = %turn_id,
                                    "runtime helper asr_realtime_session_started"
                                );
                                realtime_asr_upload = Some(upload);
                            }
                            Err(error) if turn_bridge_config.asr_realtime_enabled => {
                                warn!(
                                    call_id = %call_id,
                                    trace_id = %trace_id,
                                    turn_id = %turn_id,
                                    error = %safe_error(&error.to_string()),
                                    "runtime helper asr_realtime_start_failed_fallback"
                                );
                            }
                            Err(_) => {}
                        }
                    } else if was_in_speech {
                    if was_in_speech {
                        let push_failed = realtime_asr_upload
                            .as_mut()
                            .and_then(|upload| upload.push_48k_samples(frame.data.as_ref()).err());
@@ -2738,6 +3173,174 @@
            "runtime helper user_audio_stream_ended"
        );
    })
}

fn observe_frame_and_start_session<F>(
    vad: &mut SimpleVad,
    call_id: &str,
    trace_id: &str,
    participant_alias: &str,
    track_sid_alias: &str,
    frame_count: u64,
    elapsed_ms: u64,
    frame: &AudioFrame<'_>,
    http: Client,
    config: RealtimeAsrConfig,
    read_attributes: F,
    upload_slot: &mut Option<RealtimeAsrUpload>,
    last_fixture_sequence: &mut Option<String>,
    realtime_enabled: bool,
) -> (bool, bool, Option<FinishedSpeechTurn>)
where
    F: FnOnce() -> std::collections::HashMap<String, String>,
{
    let was_in_speech = vad.in_speech;
    let turn = vad.observe_frame(
        call_id,
        trace_id,
        participant_alias,
        track_sid_alias,
        frame_count,
        elapsed_ms,
        frame,
    );
    let is_in_speech = vad.in_speech;
    if !was_in_speech && is_in_speech {
        start_realtime_session_for_new_speech(
            http,
            config,
            call_id,
            trace_id,
            vad,
            read_attributes,
            upload_slot,
            last_fixture_sequence,
            realtime_enabled,
        );
    }
    (was_in_speech, is_in_speech, turn)
}

fn observe_bound_participant_frame<F>(
    participant_identity: &str,
    expected_participant: Option<&str>,
    vad: &mut SimpleVad,
    call_id: &str,
    trace_id: &str,
    participant_alias: &str,
    track_sid_alias: &str,
    frame_count: u64,
    elapsed_ms: u64,
    frame: &AudioFrame<'_>,
    http: Client,
    config: RealtimeAsrConfig,
    read_attributes: F,
    upload_slot: &mut Option<RealtimeAsrUpload>,
    last_fixture_sequence: &mut Option<String>,
    realtime_enabled: bool,
) -> (bool, bool, Option<FinishedSpeechTurn>)
where
    F: FnOnce() -> std::collections::HashMap<String, String>,
{
    if !is_bound_user_participant(participant_identity, expected_participant) {
        warn!(
            "audioIngressOriginStatus" = "wrong_participant_or_track",
            "runtime helper rejected audio participant before VAD/session"
        );
        return (vad.in_speech, vad.in_speech, None);
    }
    observe_frame_and_start_session(
        vad,
        call_id,
        trace_id,
        participant_alias,
        track_sid_alias,
        frame_count,
        elapsed_ms,
        frame,
        http,
        config,
        read_attributes,
        upload_slot,
        last_fixture_sequence,
        realtime_enabled,
    )
}

fn start_realtime_session_for_new_speech(
    http: Client,
    config: RealtimeAsrConfig,
    call_id: &str,
    trace_id: &str,
    vad: &SimpleVad,
    read_attributes: impl FnOnce() -> std::collections::HashMap<String, String>,
    upload_slot: &mut Option<RealtimeAsrUpload>,
    last_fixture_sequence: &mut Option<String>,
    realtime_enabled: bool,
) {
    let turn_id = format!("turn-{:04}", vad.turn_index);
    let attributes = read_attributes();
    let origin_status = AudioIngressMetadata::origin_status(&attributes);
    let metadata = match AudioIngressMetadata::from_participant(&attributes) {
        Ok(metadata) => metadata,
        Err(reason) => {
            warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                reason, audioIngressOriginStatus = %AudioIngressMetadata::rejection_status(reason),
                "runtime helper asr_realtime_metadata_rejected");
            return;
        }
    };
    if let Some(metadata) = metadata.as_ref() {
        if !fixture_sequence_is_new(
            last_fixture_sequence.as_deref(),
            &metadata.client_fixture_sequence,
        ) {
            warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                audioIngressOriginStatus = "sequence_replayed_or_regressed",
                "runtime helper asr_realtime_metadata_sequence_rejected");
            return;
        }
    }
    match RealtimeAsrUpload::start(
        http,
        config,
        call_id,
        trace_id,
        &turn_id,
        &vad.speech_samples,
        metadata.as_ref(),
    ) {
        Ok(upload) => {
            if let Some(metadata) = metadata {
                *last_fixture_sequence = Some(metadata.client_fixture_sequence);
            }
            info!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                origin_status, "runtime helper asr_realtime_session_started");
            *upload_slot = Some(upload);
        }
        Err(error) if realtime_enabled => {
            warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
                error = %safe_error(&error.to_string()),
                "runtime helper asr_realtime_start_failed_fallback");
        }
        Err(_) => {}
    }
}

fn fixture_sequence_is_new(previous: Option<&str>, current: &str) -> bool {
    let Some(previous) = previous else {
        return true;
    };
    let current_number = current
        .rsplit_once('-')
        .and_then(|(_, value)| value.parse::<u64>().ok());
    let previous_number = previous
        .rsplit_once('-')
        .and_then(|(_, value)| value.parse::<u64>().ok());
    match (previous_number, current_number) {
        (Some(previous), Some(current)) => current > previous,
        _ => previous != current,
    }
}

struct DrainedUserAudioFrame {
@@ -3489,3 +4092,512 @@
        Ok(())
    }
}

#[cfg(test)]
mod tests {
    use super::{
        ReplyChunkMarker, ReplyChunkMarkerState, RuntimeTurnDeviceOutput, RuntimeTurnStreamEvent,
        RuntimeTurnStreamState, RuntimeTurnStreamTimingPhase, runtime_session_nonce_hash,
        should_publish_device_output,
    };
    use std::{
        collections::HashSet,
        io::{Read, Write},
        net::TcpListener,
        sync::{
            Arc,
            atomic::{AtomicUsize, Ordering},
            mpsc,
        },
        thread,
        time::Duration,
    };

    #[test]
    fn production_vad_session_boundary_reads_updated_attributes() {
        let config = SimpleVadConfig {
            rms_threshold: 0.001,
            peak_threshold: 0.01,
            start_frames: 2,
            end_silence_ms: 100,
            min_speech_ms: 1,
            max_turn_ms: 1_000,
            initial_ignore_ms: 0,
        };
        let mut vad = SimpleVad::new(config);
        let samples = vec![1_000i16; 160];
        let frame = AudioFrame {
            data: samples.as_slice().into(),
            sample_rate: 16_000,
            num_channels: 1,
            samples_per_channel: 160,
        };
        let mut attributes = std::collections::HashMap::from([
            (
                "inputSourceCategory".to_string(),
                "controlled_fixture".to_string(),
            ),
            (
                "clientFixtureSequence".to_string(),
                "fixture-01".to_string(),
            ),
        ]);
        let mut starts = Vec::new();
        for (session_index, sequence) in [(1, "fixture-01"), (2, "fixture-02")] {
            let was_in_speech = vad.in_speech;
            vad.observe_frame(
                "call-001",
                "trace-001",
                "participant",
                "track",
                session_index * 2 - 1,
                1_000 * session_index,
                &frame,
            );
            vad.observe_frame(
                "call-001",
                "trace-001",
                "participant",
                "track",
                session_index * 2,
                1_000 * session_index + 10,
                &frame,
            );
            let is_in_speech = vad.in_speech;
            assert!(!was_in_speech && is_in_speech);
            attributes.insert("clientFixtureSequence".to_string(), sequence.to_string());
            let metadata = AudioIngressMetadata::from_participant(&attributes)
                .expect("valid participant attributes")
                .expect("controlled fixture metadata");
            let session_line = asr_realtime::session_start_line(
                "call-001",
                "trace-001",
                &format!("turn-{session_index:04}"),
                "nonce-001",
                Some(&metadata),
            )
            .expect("session start line");
            let session_json: serde_json::Value =
                serde_json::from_slice(&session_line).expect("session start json");
            assert_eq!(sequence, session_json["clientFixtureSequence"]);
            starts.push(metadata.client_fixture_sequence);
            vad.reset_current_turn();
        }
        assert_eq!(vec!["fixture-01", "fixture-02"], starts);
        attributes.insert("inputSourceCategory".to_string(), "other".to_string());
        assert!(AudioIngressMetadata::from_participant(&attributes).is_err());
        assert!(
            AudioIngressMetadata::from_participant(&std::collections::HashMap::new())
                .expect("missing attributes is absent")
                .is_none()
        );
        attributes.insert(
            "clientFixtureSequence".to_string(),
            "fixture-01".to_string(),
        );
        assert!(AudioIngressMetadata::from_participant(&attributes).is_err());
    }

    #[test]
    fn production_observer_rejects_wrong_participant_before_vad_session() {
        assert!(!is_bound_user_participant(
            "participant-other",
            Some("participant-user")
        ));
        assert!(is_bound_user_participant(
            "participant-user",
            Some("participant-user")
        ));
        assert!(is_bound_user_participant("participant-any", None));
    }

    #[tokio::test]
    async fn production_observer_vad_to_session_entry_reads_each_updated_attribute() {
        let listener = TcpListener::bind("127.0.0.1:0").expect("bind local ASR fixture");
        let address = listener.local_addr().expect("fixture address");
        let (request_tx, request_rx) = mpsc::channel::<String>();
        let captured_count = Arc::new(AtomicUsize::new(0));
        let captured_count_for_server = Arc::clone(&captured_count);
        let server = thread::spawn(move || {
            for _ in 0..2 {
                let (mut stream, _) = listener.accept().expect("accept ASR session");
                stream
                    .set_read_timeout(Some(Duration::from_secs(2)))
                    .expect("set fixture timeout");
                let mut bytes = Vec::new();
                let mut buffer = [0_u8; 4096];
                loop {
                    match stream.read(&mut buffer) {
                        Ok(0) => break,
                        Ok(size) => {
                            bytes.extend_from_slice(&buffer[..size]);
                            if bytes.windows(7).any(|window| window == b"0\r\n\r\n") {
                                break;
                            }
                        }
                        Err(_) => break,
                    }
                }
                request_tx
                    .send(String::from_utf8_lossy(&bytes).into_owned())
                    .expect("capture ASR request");
                captured_count_for_server.fetch_add(1, Ordering::SeqCst);
                stream
                    .write_all(b"HTTP/1.1 200 OK\r\ncontent-type: application/json\r\ncontent-length: 39\r\nconnection: close\r\n\r\n{\"code\":0,\"data\":{\"status\":\"ok\"}}")
                    .expect("write fixture response");
            }
        });
        let mut vad = SimpleVad::new(SimpleVadConfig {
            rms_threshold: 0.001,
            peak_threshold: 0.01,
            start_frames: 1,
            end_silence_ms: 100,
            min_speech_ms: 1,
            max_turn_ms: 1_000,
            initial_ignore_ms: 0,
        });
        let frame_data = vec![1_000i16; 160];
        let frame = AudioFrame {
            data: frame_data.as_slice().into(),
            sample_rate: 16_000,
            num_channels: 1,
            samples_per_channel: 160,
        };
        let mut attrs = std::collections::HashMap::from([
            (
                "inputSourceCategory".to_string(),
                "controlled_fixture".to_string(),
            ),
            (
                "clientFixtureSequence".to_string(),
                "fixture-01".to_string(),
            ),
        ]);
        let mut upload = None;
        let mut last_fixture_sequence = None;
        let config = RealtimeAsrConfig {
            enabled: true,
            url: Some(format!("http://{address}/runtime/asr/realtime")),
            runtime_token: Some("test".to_string()),
            runtime_session_nonce: Some("test".to_string()),
            chunk_duration_ms: 200,
        };
        let (was, is, turn) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            1,
            1_000,
            &frame,
            Client::new(),
            config.clone(),
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(!was && is && turn.is_none());
        assert!(upload.is_some());
        upload.take().unwrap().cancel("test").await;
        vad.reset_current_turn();
        attrs.insert(
            "clientFixtureSequence".to_string(),
            "fixture-02".to_string(),
        );
        let (was, is, turn) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            2,
            2_000,
            &frame,
            Client::new(),
            config,
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(!was && is && turn.is_none());
        assert!(upload.is_some());
        upload.take().unwrap().cancel("test").await;

        let first_request = request_rx
            .recv_timeout(Duration::from_secs(2))
            .expect("first session request");
        let second_request = request_rx
            .recv_timeout(Duration::from_secs(2))
            .expect("second session request");
        assert!(first_request.contains("\\\"clientFixtureSequence\\\":\\\"fixture-01\\\""));
        assert!(second_request.contains("\\\"clientFixtureSequence\\\":\\\"fixture-02\\\""));
        assert!(
            first_request
                .contains("\\\"audioIngressOriginStatus\\\":\\\"controlled_fixture_bound\\\"")
        );
        assert!(
            second_request
                .contains("\\\"audioIngressOriginStatus\\\":\\\"controlled_fixture_bound\\\"")
        );
        // The same production boundary rejects a wrong participant before VAD/session creation.
        assert!(!is_bound_user_participant(
            "participant-other",
            Some("participant-user")
        ));
        assert_eq!(0, request_rx.try_iter().count());
        vad.reset_current_turn();
        attrs.insert(
            "clientFixtureSequence".to_string(),
            "fixture-03".to_string(),
        );
        let (_, is_wrong, wrong_turn) = observe_bound_participant_frame(
            "participant-other",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            3,
            3_000,
            &frame,
            Client::new(),
            RealtimeAsrConfig {
                enabled: true,
                url: Some(format!("http://{address}/runtime/asr/realtime")),
                runtime_token: Some("test".to_string()),
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(!is_wrong && wrong_turn.is_none() && upload.is_none());
        vad.reset_current_turn();
        attrs.insert(
            "clientFixtureSequence".to_string(),
            "fixture-01".to_string(),
        );
        let (_, _, _) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            4,
            4_000,
            &frame,
            Client::new(),
            RealtimeAsrConfig {
                enabled: true,
                url: Some(format!("http://{address}/runtime/asr/realtime")),
                runtime_token: Some("test".to_string()),
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(upload.is_none());
        vad.reset_current_turn();
        let (_, _, _) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-user",
            "track-001",
            5,
            5_000,
            &frame,
            Client::new(),
            RealtimeAsrConfig {
                enabled: true,
                url: Some(format!("http://{address}/runtime/asr/realtime")),
                runtime_token: Some("test".to_string()),
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(upload.is_none());
        vad.reset_current_turn();
        attrs.insert("inputSourceCategory".to_string(), "other".to_string());
        let mut invalid_upload = None;
        let (_, _, invalid_turn) = observe_bound_participant_frame(
            "participant-user",
            Some("participant-user"),
            &mut vad,
            "call-001",
            "trace-001",
            "participant-other",
            "track-001",
            3,
            3_000,
            &frame,
            Client::new(),
            RealtimeAsrConfig {
                enabled: true,
                url: Some("http://127.0.0.1:9".to_string()),
                runtime_token: Some("test".to_string()),
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            || attrs,
            &mut invalid_upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(invalid_turn.is_none());
        assert!(invalid_upload.is_none());
        assert_eq!(2, captured_count.load(Ordering::SeqCst));
        assert_eq!(0, request_rx.try_iter().count());
        server.join().expect("fixture server");
    }

    #[test]
    fn reply_chunk_marker_state_emits_turn_first_once_and_later_segment_first_once() {
        let mut state = ReplyChunkMarkerState::default();

        assert_eq!(ReplyChunkMarker::FirstReply, state.observe(Some(1)));
        assert_eq!(ReplyChunkMarker::None, state.observe(Some(1)));
        assert_eq!(ReplyChunkMarker::SegmentFirst, state.observe(Some(2)));
        assert_eq!(ReplyChunkMarker::None, state.observe(Some(2)));
        assert_eq!(ReplyChunkMarker::SegmentFirst, state.observe(Some(3)));
    }

    #[test]
    fn reply_chunk_marker_state_without_segment_only_emits_turn_first() {
        let mut state = ReplyChunkMarkerState::default();

        assert_eq!(ReplyChunkMarker::FirstReply, state.observe(None));
        assert_eq!(ReplyChunkMarker::None, state.observe(None));
    }

    #[test]
    fn runtime_turn_stream_audio_chunk_reads_segment_seq() {
        let event: RuntimeTurnStreamEvent = serde_json::from_str(
            r#"{"type":"reply_audio_chunk","audioChunk":{"chunkSeq":4,"segmentSeq":2,"format":"pcm_s16le","payloadBase64":"AA==","last":false}}"#,
        )
        .expect("turn stream event");

        assert_eq!(
            Some(2),
            event.audio_chunk.and_then(|chunk| chunk.segment_seq)
        );
    }

    #[test]
    fn runtime_turn_stream_reads_frozen_m5_and_audio_chunk_timing_contract() {
        let activity: RuntimeTurnStreamEvent = serde_json::from_str(
            r#"{"type":"activity","callId":"call-1","traceId":"trace-1","turnId":"turn-1","activity":{"eventType":"tts_first_audio_chunk_ready","extension":{"streamTimingVersion":1,"streamAnchorId":"0123456789abcdef","streamAnchorServerDeltaMs":1200,"runtimeSessionNonceHash":"abcdef012345","segmentSeq":1,"streamTimingValidation":"bound"}}}"#,
        )
        .expect("m5 activity event");
        let audio: RuntimeTurnStreamEvent = serde_json::from_str(
            r#"{"type":"reply_audio_chunk","callId":"call-1","traceId":"trace-1","turnId":"turn-1","audioChunk":{"chunkSeq":1,"segmentSeq":1,"streamTimingVersion":1,"streamAnchorId":"0123456789abcdef","format":"pcm_s16le","payloadBase64":"AA==","last":false}}"#,
        )
        .expect("timed audio chunk event");

        assert_eq!(Some("call-1"), activity.call_id.as_deref());
        assert_eq!(Some("trace-1"), activity.trace_id.as_deref());
        assert_eq!(Some("turn-1"), activity.turn_id.as_deref());
        let extension = activity.activity.unwrap().extension.unwrap();
        assert_eq!(Some(1), extension.stream_timing_version);
        assert_eq!(Some(1200), extension.stream_anchor_server_delta_ms);
        assert_eq!(
            Some("0123456789abcdef"),
            audio.audio_chunk.unwrap().stream_anchor_id.as_deref()
        );
    }

    #[test]
    fn runtime_turn_stream_timing_records_m6_and_m7_once_then_rejects_terminal_late_events() {
        let nonce = "runtime-nonce";
        let nonce_hash = runtime_session_nonce_hash(nonce);
        let activity: RuntimeTurnStreamEvent = serde_json::from_str(&format!(
            r#"{{"type":"activity","callId":"call-1","traceId":"trace-1","turnId":"turn-1","activity":{{"eventType":"tts_first_audio_chunk_ready","extension":{{"streamTimingVersion":1,"streamAnchorId":"0123456789abcdef","streamAnchorServerDeltaMs":1200,"runtimeSessionNonceHash":"{nonce_hash}","segmentSeq":1,"streamTimingValidation":"bound"}}}}}}"#,
        ))
        .expect("m5 activity event");
        let audio: RuntimeTurnStreamEvent = serde_json::from_str(
            r#"{"type":"reply_audio_chunk","callId":"call-1","traceId":"trace-1","turnId":"turn-1","audioChunk":{"chunkSeq":1,"segmentSeq":1,"streamTimingVersion":1,"streamAnchorId":"0123456789abcdef","format":"pcm_s16le","payloadBase64":"AA==","last":false}}"#,
        )
        .expect("timed audio chunk event");
        let mut state = RuntimeTurnStreamState::default();

        assert!(state.arm_timing_anchor("call-1", "trace-1", "turn-1", nonce, &activity));
        let chunk = audio.audio_chunk.as_ref().expect("audio chunk");
        assert!(
            state
                .record_m6("call-1", "trace-1", "turn-1", &audio, chunk)
                .is_some()
        );
        assert!(
            state
                .record_m6("call-1", "trace-1", "turn-1", &audio, chunk)
                .is_none()
        );
        assert!(state.record_m7().is_some());
        assert!(state.record_m7().is_none());

        state.close_timing();
        assert_eq!(RuntimeTurnStreamTimingPhase::Closed, state.timing.phase);
        assert!(state.timing.anchor.is_none());
        assert!(!state.arm_timing_anchor("call-1", "trace-1", "turn-1", nonce, &activity));
        assert!(
            state
                .record_m6("call-1", "trace-1", "turn-1", &audio, chunk)
                .is_none()
        );
        assert!(state.record_m7().is_none());
    }

    #[test]
    fn device_output_contract_is_reliable_and_deduplicated() {
        let output = RuntimeTurnDeviceOutput {
            command_id: Some("cmd-1".to_string()),
            command_code: Some("custom.app.DeviceLevelChange".to_string()),
            params: Some(serde_json::json!({"level": 1})),
        };
        let mut published = HashSet::new();
        assert!(should_publish_device_output(&mut published, &output));
        assert!(!should_publish_device_output(&mut published, &output));
        assert_eq!(published.len(), 1);
    }

    #[test]
    fn device_output_invalid_or_missing_command_is_fail_closed() {
        for output in [
            RuntimeTurnDeviceOutput {
                command_id: None,
                command_code: Some("custom.app.DeviceLevelChange".to_string()),
                params: None,
            },
            RuntimeTurnDeviceOutput {
                command_id: Some("cmd-1".to_string()),
                command_code: None,
                params: None,
            },
        ] {
            let mut published = HashSet::new();
            assert!(!should_publish_device_output(&mut published, &output));
            assert!(published.is_empty());
        }
    }
}