cai
2026-08-08 2fe34384cd7519f1aebb5f3ac818fe008541e0a6
src/main.rs
@@ -3000,6 +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;
@@ -3033,8 +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,
@@ -3042,21 +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 {
                        start_realtime_session_for_new_speech(
                            http.clone(),
                            turn_bridge_config.realtime_asr_config(),
                            &call_id,
                            &trace_id,
                            vad,
                            || participant.attributes(),
                            &mut realtime_asr_upload,
                            turn_bridge_config.asr_realtime_enabled,
                        );
                    } 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());
@@ -3175,6 +3170,52 @@
    })
}
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,
@@ -3183,19 +3224,41 @@
    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);
    match RealtimeAsrUpload::start_with_participant_attributes(
    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,
        read_attributes,
        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);
@@ -3206,6 +3269,22 @@
                "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,
    }
}
@@ -3966,7 +4045,14 @@
        RuntimeTurnStreamState, RuntimeTurnStreamTimingPhase, runtime_session_nonce_hash,
        should_publish_device_output,
    };
    use std::collections::HashSet;
    use std::{
        collections::HashSet,
        io::{Read, Write},
        net::TcpListener,
        sync::mpsc,
        thread,
        time::Duration,
    };
    #[test]
    fn production_vad_session_boundary_reads_updated_attributes() {
@@ -4068,6 +4154,37 @@
    #[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,
@@ -4095,18 +4212,142 @@
            ),
        ]);
        let mut upload = None;
        let was = vad.in_speech;
        vad.observe_frame(
        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",
            "track",
            "participant-user",
            "track-001",
            1,
            1_000,
            &frame,
            Client::new(),
            config.clone(),
            || attrs.clone(),
            &mut upload,
            &mut last_fixture_sequence,
            true,
        );
        assert!(!was && vad.in_speech);
        start_realtime_session_for_new_speech(
        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,
@@ -4115,38 +4356,13 @@
                runtime_session_nonce: Some("test".to_string()),
                chunk_duration_ms: 200,
            },
            "call-001",
            "trace-001",
            &vad,
            || attrs.clone(),
            &mut upload,
            || attrs,
            &mut invalid_upload,
            &mut last_fixture_sequence,
            true,
        );
        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 = vad.in_speech;
        vad.observe_frame(
            "call-001",
            "trace-001",
            "participant",
            "track",
            2,
            2_000,
            &frame,
        );
        assert!(!was && vad.in_speech);
        assert_eq!(
            "fixture-02",
            AudioIngressMetadata::from_participant(&attrs)
                .expect("valid attributes")
                .expect("bound")
                .client_fixture_sequence
        );
        assert!(invalid_turn.is_none());
        assert!(invalid_upload.is_none());
    }
    #[test]