From c5bfc24db1ae38d8488e566a2c9ff8afd42c0e5c Mon Sep 17 00:00:00 2001
From: cai <cai@nbcai.cc>
Date: Sat, 08 Aug 2026 19:16:59 +0800
Subject: [PATCH] chore: bind final observer test candidate
---
src/main.rs | 576 +++++++++++++++++++++++++++++++++++++++++++++++++++++---
1 files changed, 538 insertions(+), 38 deletions(-)
diff --git a/src/main.rs b/src/main.rs
index 611993a..29c1d2a 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -15,7 +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;
@@ -28,7 +30,7 @@
options::TrackPublishOptions,
prelude::{
DataPacket, LocalAudioTrack, LocalTrack, ParticipantIdentity, RemoteAudioTrack,
- RemoteTrack, Room, RoomEvent, RoomOptions,
+ RemoteParticipant, RemoteTrack, Room, RoomEvent, RoomOptions,
},
};
use reqwest::Client;
@@ -707,6 +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 {
@@ -728,9 +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(
@@ -743,6 +751,7 @@
turn_bridge_config: TurnBridgeConfig,
http: Client,
sink: Arc<BotAudioOutputSink>,
+ user_participant_identity: Option<String>,
) {
info!(
call_id = %call_id,
@@ -766,6 +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();
@@ -791,6 +809,7 @@
turn_bridge_config.clone(),
http.clone(),
sink.clone(),
+ participant,
);
}
RoomEvent::TrackSubscribed {
@@ -985,7 +1004,11 @@
.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
@@ -1593,6 +1616,9 @@
}
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);
@@ -2421,6 +2447,7 @@
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,
@@ -2438,6 +2465,7 @@
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,
@@ -2830,6 +2858,30 @@
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<()> {
if value.is_empty()
|| value.contains('/')
@@ -2869,6 +2921,7 @@
turn_bridge_config: TurnBridgeConfig,
http: Client,
sink: Arc<BotAudioOutputSink>,
+ participant: RemoteParticipant,
) -> JoinHandle<()> {
tokio::spawn(async move {
let mut stream = NativeAudioStream::new(
@@ -2947,6 +3000,7 @@
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;
@@ -2980,8 +3034,8 @@
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 (was_in_speech, is_in_speech, turn) = observe_frame_and_start_session(
+ vad,
&call_id,
&trace_id,
&participant_alias,
@@ -2989,40 +3043,15 @@
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());
@@ -3139,6 +3168,124 @@
"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 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 metadata = match AudioIngressMetadata::from_participant(&read_attributes()) {
+ Ok(metadata) => metadata,
+ Err(reason) => {
+ warn!(call_id = %call_id, trace_id = %trace_id, turn_id = %turn_id,
+ 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,
+ "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,
+ "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 {
@@ -3894,9 +4041,329 @@
#[cfg(test)]
mod tests {
use super::{
- ReplyChunkMarker, ReplyChunkMarkerState, RuntimeTurnStreamEvent, RuntimeTurnStreamState,
- RuntimeTurnStreamTimingPhase, runtime_session_nonce_hash,
+ ReplyChunkMarker, ReplyChunkMarkerState, RuntimeTurnDeviceOutput, RuntimeTurnStreamEvent,
+ RuntimeTurnStreamState, RuntimeTurnStreamTimingPhase, runtime_session_nonce_hash,
+ should_publish_device_output,
};
+ use std::{
+ collections::HashSet,
+ io::{Read, Write},
+ net::TcpListener,
+ sync::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 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");
+ 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_frame_and_start_session(
+ &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_frame_and_start_session(
+ &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\\\""));
+ server.join().expect("fixture server");
+
+ // 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-01".to_string(),
+ );
+ let (_, _, _) = observe_frame_and_start_session(
+ &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!(upload.is_none());
+ vad.reset_current_turn();
+ attrs.insert(
+ "clientFixtureSequence".to_string(),
+ "fixture-00".to_string(),
+ );
+ let (_, _, _) = observe_frame_and_start_session(
+ &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();
+ attrs.insert("inputSourceCategory".to_string(), "other".to_string());
+ let mut invalid_upload = None;
+ let (_, _, invalid_turn) = observe_frame_and_start_session(
+ &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());
+ }
#[test]
fn reply_chunk_marker_state_emits_turn_first_once_and_later_segment_first_once() {
@@ -3993,4 +4460,37 @@
);
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());
+ }
+ }
}
--
Gitblit v1.9.3