From 3c5a17954a45d279295059da078aefcf956568de Mon Sep 17 00:00:00 2001
From: cai <cai@nbcai.cc>
Date: Sat, 08 Aug 2026 19:07:38 +0800
Subject: [PATCH] chore: bind final observer test candidate

---
 src/main.rs |  270 ++++++++++++++++++++++++++++++++++++++++++++++++-----
 1 files changed, 243 insertions(+), 27 deletions(-)

diff --git a/src/main.rs b/src/main.rs
index e76d222..4ab9a13 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 {
@@ -2902,6 +2921,7 @@
     turn_bridge_config: TurnBridgeConfig,
     http: Client,
     sink: Arc<BotAudioOutputSink>,
+    participant: RemoteParticipant,
 ) -> JoinHandle<()> {
     tokio::spawn(async move {
         let mut stream = NativeAudioStream::new(
@@ -3026,35 +3046,16 @@
                     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(
+                        start_realtime_session_for_new_speech(
                             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(_) => {}
-                        }
+                            vad,
+                            || participant.attributes(),
+                            &mut realtime_asr_upload,
+                            turn_bridge_config.asr_realtime_enabled,
+                        );
                     } else if was_in_speech {
                         let push_failed = realtime_asr_upload
                             .as_mut()
@@ -3172,6 +3173,40 @@
             "runtime helper user_audio_stream_ended"
         );
     })
+}
+
+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>,
+    realtime_enabled: bool,
+) {
+    let turn_id = format!("turn-{:04}", vad.turn_index);
+    match RealtimeAsrUpload::start_with_participant_attributes(
+        http,
+        config,
+        call_id,
+        trace_id,
+        &turn_id,
+        &vad.speech_samples,
+        read_attributes,
+    ) {
+        Ok(upload) => {
+            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(_) => {}
+    }
 }
 
 struct DrainedUserAudioFrame {
@@ -3934,6 +3969,187 @@
     use std::collections::HashSet;
 
     #[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 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 was = vad.in_speech;
+        vad.observe_frame(
+            "call-001",
+            "trace-001",
+            "participant",
+            "track",
+            1,
+            1_000,
+            &frame,
+        );
+        assert!(!was && vad.in_speech);
+        start_realtime_session_for_new_speech(
+            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,
+            },
+            "call-001",
+            "trace-001",
+            &vad,
+            || attrs.clone(),
+            &mut upload,
+            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
+        );
+    }
+
+    #[test]
     fn reply_chunk_marker_state_emits_turn_first_once_and_later_segment_first_once() {
         let mut state = ReplyChunkMarkerState::default();
 

--
Gitblit v1.9.3