Ybr8u speaker indentification

app/lib/backend/http/api/users.dart

@@ -243,6 +243,18 @@ Future<bool> deletePerson(String personId) async {
   return response.statusCode == 204;
 }
 
+Future<bool> deletePersonSpeechSample(String personId, int sampleIndex) async {
+  var response = await makeApiCall(
+    url: '${Env.apiBaseUrl}v1/users/people/$personId/speech-samples/$sampleIndex',
+    headers: {},
+    method: 'DELETE',
+    body: '',
+  );
+  if (response == null) return false;
+  debugPrint('deletePersonSpeechSample response: ${response.body}');
+  return response.statusCode == 200;
+}
+
 Future<String> getFollowUpQuestion({String conversationId = '0'}) async {
   var response = await makeApiCall(
     url: '${Env.apiBaseUrl}v1/joan/$conversationId/followup-question',

app/lib/pages/settings/people.dart

@@ -8,7 +8,6 @@ import 'package:omi/providers/people_provider.dart';
 import 'package:omi/providers/connectivity_provider.dart';
 import 'package:omi/widgets/dialog.dart';
 import 'package:omi/widgets/extensions/functions.dart';
-import 'package:just_audio/just_audio.dart';
 import 'package:omi/utils/l10n_extensions.dart';
 import 'package:provider/provider.dart';
 
@@ -161,7 +160,7 @@ class _UserPeoplePageState extends State<_UserPeoplePage> {
     );
   }
 
-  Future<void> _confirmDeleteSample(int peopleIdx, Person person, String url, PeopleProvider provider) async {
+  Future<void> _confirmDeleteSample(int peopleIdx, Person person, int sampleIdx, PeopleProvider provider) async {
     final connectivityProvider = Provider.of<ConnectivityProvider>(context, listen: false);
     if (!connectivityProvider.isConnected) {
       ConnectivityProvider.showNoInternetDialog(context);
@@ -180,7 +179,7 @@ class _UserPeoplePageState extends State<_UserPeoplePage> {
     );
 
     if (confirmed == true) {
-      provider.deletePersonSample(peopleIdx, url);
+      await provider.deletePersonSample(peopleIdx, sampleIdx);
     }
   }
 
@@ -297,20 +296,11 @@ class _UserPeoplePageState extends State<_UserPeoplePage> {
                                             ),
                                             onPressed: () => provider.playPause(index, j, sample),
                                           ),
-                                          title: Text(index == 0
+                                          title: Text(j == 0
                                               ? context.l10n.speechProfile
-                                              : context.l10n.sampleNumber(index)),
-                                          onTap: () => _confirmDeleteSample(index, person, sample, provider),
-                                          subtitle: FutureBuilder<Duration?>(
-                                            future: AudioPlayer().setUrl(sample),
-                                            builder: (context, snapshot) {
-                                              if (snapshot.hasData) {
-                                                return Text(context.l10n.secondsCount(snapshot.data!.inSeconds));
-                                              } else {
-                                                return Text(context.l10n.loadingDuration);
-                                              }
-                                            },
-                                          ),
+                                              : context.l10n.sampleNumber(j)),
+                                          onTap: () => _confirmDeleteSample(index, person, j, provider),
+                                          subtitle: Text('Tap to delete'),
                                         )),
                                   ],
                                 ),

app/lib/providers/people_provider.dart

@@ -1,5 +1,4 @@
 import 'package:flutter/cupertino.dart';
-import 'package:omi/backend/http/api/speech_profile.dart';
 import 'package:omi/backend/http/api/users.dart';
 import 'package:omi/backend/preferences.dart';
 import 'package:omi/backend/schema/person.dart';
@@ -106,21 +105,17 @@ class PeopleProvider extends BaseProvider {
     notifyListeners();
   }
 
-  String _getFileNameFromUrl(String url) {
-    Uri uri = Uri.parse(url);
-    String fileName = uri.pathSegments.last;
-    return fileName.split('.').first;
-  }
+  Future<void> deletePersonSample(int personIdx, int sampleIdx) async {
+    String personId = people[personIdx].id;
 
-  void deletePersonSample(int personIdx, String url) {
-    String name = _getFileNameFromUrl(url);
-    var parts = name.split('_segment_');
-    String conversationId = parts[0];
-    int segmentIdx = int.parse(parts[1]);
-    deleteProfileSample(conversationId, segmentIdx, personId: people[personIdx].id);
-    people[personIdx].speechSamples!.remove(url);
-    SharedPreferencesUtil().replaceCachedPerson(people[personIdx]);
-    notifyListeners();
+    bool success = await deletePersonSpeechSample(personId, sampleIdx);
+    if (success) {
+      people[personIdx].speechSamples!.removeAt(sampleIdx);
+      SharedPreferencesUtil().replaceCachedPerson(people[personIdx]);
+      notifyListeners();
+    } else {
+      debugPrint('Failed to delete speech sample at index: $sampleIdx');
+    }
   }
 
   void deletePersonProvider(Person person) {

backend/charts/backend-listen/dev_omi_backend_listen_values.yaml

@@ -114,6 +114,8 @@ env:
     value: "http://34.172.155.20:80/v1/vad"
   - name: HOSTED_SPEECH_PROFILE_API_URL
     value: "http://34.172.155.20:80/v1/speaker-identification"
+  - name: HOSTED_SPEAKER_EMBEDDING_API_URL
+    value: "http://34.172.155.20:80"
   - name: PINECONE_API_KEY
     valueFrom:
       secretKeyRef:

backend/charts/backend-listen/prod_omi_backend_listen_values.yaml

@@ -107,6 +107,8 @@ env:
     value: "http://172.16.128.101:8080/v1/vad"
   - name: HOSTED_SPEECH_PROFILE_API_URL
     value: "http://172.16.128.101:8080/v1/speaker-identification"
+  - name: HOSTED_SPEAKER_EMBEDDING_API_URL
+    value: "http://diarizer.omi.me:80"
   - name: PINECONE_API_KEY
     valueFrom:
       secretKeyRef:

backend/charts/pusher/dev_omi_pusher_values.yaml

@@ -106,6 +106,8 @@ env:
     value: "http://34.172.155.20:80/v1/vad"
   - name: HOSTED_SPEECH_PROFILE_API_URL
     value: "http://34.172.155.20:80/v1/speaker-identification"
+  - name: HOSTED_SPEAKER_EMBEDDING_API_URL
+    value: "http://34.172.155.20:80"
   - name: PINECONE_API_KEY
     valueFrom:
       secretKeyRef:

backend/charts/pusher/prod_omi_pusher_values.yaml

@@ -111,6 +111,8 @@ env:
     value: "http://vad.omi.me:80/v1/vad"
   - name: HOSTED_SPEECH_PROFILE_API_URL
     value: "http://vad.omi.me:80/v1/speaker-identification"
+  - name: HOSTED_SPEAKER_EMBEDDING_API_URL
+    value: "http://diarizer.omi.me:80"
   - name: PINECONE_API_KEY
     valueFrom:
       secretKeyRef:

backend/database/users.py

@@ -100,6 +100,127 @@ def delete_person(uid: str, person_id: str):
     person_ref.delete()
 
 
+def add_person_speech_sample(uid: str, person_id: str, sample_path: str, max_samples: int = 5) -> bool:
+    """
+    Append speech sample path to person's speech_samples list.
+    Limits to max_samples to prevent unlimited growth.
+
+    Args:
+        uid: User ID
+        person_id: Person ID
+        sample_path: GCS path to the speech sample
+        max_samples: Maximum number of samples to keep (default 5)
+
+    Returns:
+        True if sample was added, False if limit reached
+    """
+    person_ref = db.collection('users').document(uid).collection('people').document(person_id)
+    person_doc = person_ref.get()
+
+    if not person_doc.exists:
+        return False
+
+    person_data = person_doc.to_dict()
+    current_samples = person_data.get('speech_samples', [])
+
+    # Check if we've hit the limit
+    if len(current_samples) >= max_samples:
+        return False
+
+    person_ref.update(
+        {
+            'speech_samples': firestore.ArrayUnion([sample_path]),
+            'updated_at': datetime.now(timezone.utc),
+        }
+    )
+    return True
+
+
+def get_person_speech_samples_count(uid: str, person_id: str) -> int:
+    """Get the count of speech samples for a person."""
+    person_ref = db.collection('users').document(uid).collection('people').document(person_id)
+    person_doc = person_ref.get()
+
+    if not person_doc.exists:
+        return 0
+
+    person_data = person_doc.to_dict()
+    return len(person_data.get('speech_samples', []))
+
+
+def remove_person_speech_sample(uid: str, person_id: str, sample_path: str) -> bool:
+    """
+    Remove a speech sample path from person's speech_samples list.
+
+    Args:
+        uid: User ID
+        person_id: Person ID
+        sample_path: GCS path to remove
+
+    Returns:
+        True if removed, False if person not found
+    """
+    person_ref = db.collection('users').document(uid).collection('people').document(person_id)
+    person_doc = person_ref.get()
+
+    if not person_doc.exists:
+        return False
+
+    person_ref.update({
+        'speech_samples': firestore.ArrayRemove([sample_path]),
+        'updated_at': datetime.now(timezone.utc),
+    })
+    return True
+
+
+def set_person_speaker_embedding(uid: str, person_id: str, embedding: list) -> bool:
+    """
+    Store speaker embedding for a person.
+
+    Args:
+        uid: User ID
+        person_id: Person ID
+        embedding: List of floats representing the speaker embedding
+
+    Returns:
+        True if stored successfully, False if person not found
+    """
+    person_ref = db.collection('users').document(uid).collection('people').document(person_id)
+    person_doc = person_ref.get()
+
+    if not person_doc.exists:
+        return False
+
+    person_ref.update(
+        {
+            'speaker_embedding': embedding,
+            'updated_at': datetime.now(timezone.utc),
+        }
+    )
+    return True
+
+
+def get_person_speaker_embedding(uid: str, person_id: str) -> Optional[list]:
+    """
+    Get speaker embedding for a person.
+
+    Args:
+        uid: User ID
+        person_id: Person ID
+
+    Returns:
+        List of floats representing the embedding, or None if not found
+    """
+    person_ref = db.collection('users').document(uid).collection('people').document(person_id)
+    person_doc = person_ref.get()
+
+    if not person_doc.exists:
+        return None
+
+    person_data = person_doc.to_dict()
+    return person_data.get('speaker_embedding')
+
+
 def delete_user_data(uid: str):
     user_ref = db.collection('users').document(uid)
     if not user_ref.get().exists:

backend/routers/conversations.py

@@ -32,6 +32,7 @@
 from utils.conversations.process_conversation import process_conversation, retrieve_in_progress_conversation
 from utils.conversations.search import search_conversations
 from utils.llm.conversation_processing import generate_summary_with_prompt
+from utils.speaker_identification import extract_speaker_samples
 from utils.other import endpoints as auth
 from utils.other.storage import get_conversation_recording_if_exists
 from utils.app_integrations import trigger_external_integrations
@@ -495,6 +496,7 @@ def set_assignee_conversation_segment(
 def assign_segments_bulk(
     conversation_id: str,
     data: BulkAssignSegmentsRequest,
+    background_tasks: BackgroundTasks,
     uid: str = Depends(auth.get_current_user_uid),
 ):
     conversation = _get_valid_conversation_by_id(uid, conversation_id)
@@ -521,6 +523,17 @@ def assign_segments_bulk(
     conversations_db.update_conversation_segments(
         uid, conversation_id, [segment.dict() for segment in conversation.transcript_segments]
     )
+
+    # Trigger speaker sample extraction when assigning to a person
+    if data.assign_type == 'person_id' and value:
+        background_tasks.add_task(
+            extract_speaker_samples,
+            uid=uid,
+            person_id=value,
+            conversation_id=conversation_id,
+            segment_ids=data.segment_ids,
+        )
+
     return conversation