updates

livekit · Jan 24, 2025 · 400d245 · 400d245
1 parent 288a3ef
commit 400d245
Show file tree

Hide file tree

Showing 2 changed files with 29 additions and 7 deletions.
diff --git a/livekit-plugins/livekit-plugins-google/livekit/plugins/google/beta/realtime/realtime_api.py b/livekit-plugins/livekit-plugins-google/livekit/plugins/google/beta/realtime/realtime_api.py
@@ -302,7 +302,7 @@ def __init__(
             self._transcriber.on("input_speech_done", self._on_input_speech_done)
         if self._opts.enable_agent_audio_transcription:
             self._agent_transcriber = TranscriberSession(
-                client=self._client, model=self._opts.model
+                client=self._client, model=self._opts.model, sample_rate=24000
             )
             self._agent_transcriber.on("input_speech_done", self._on_agent_speech_done)
         # init dummy task

diff --git a/livekit-plugins/livekit-plugins-google/livekit/plugins/google/beta/realtime/transcriber.py b/livekit-plugins/livekit-plugins-google/livekit/plugins/google/beta/realtime/transcriber.py
@@ -49,13 +49,20 @@ def __init__(
         *,
         client: genai.Client,
         model: LiveAPIModels | str,
+        sample_rate: int = 16000,
     ):
         """
         Initializes a TranscriberSession instance for interacting with Google's Realtime API.
         """
         super().__init__()
         self._client = client
         self._model = model
+        self._sample_rate = sample_rate
+        self._bstream = utils.audio.AudioByteStream(
+            16000,
+            1,
+            samples_per_channel=1200,
+        )
         self._closed = False
         system_instructions = types.Content(
             parts=[types.Part(text=SYSTEM_INSTRUCTIONS)]
@@ -77,13 +84,28 @@ def __init__(
     def _push_audio(self, frame: rtc.AudioFrame) -> None:
         if self._closed:
             return
-        self._queue_msg(
-            types.LiveClientRealtimeInput(
-                media_chunks=[
-                    types.Blob(data=frame.data.tobytes(), mime_type="audio/pcm")
-                ]
+        if self._sample_rate != 16000:
+            bstream = utils.audio.AudioByteStream(
+                16000,
+                1,
+                samples_per_channel=1200,
+            )
+            for f in bstream.write(frame.data.tobytes()):
+                self._queue_msg(
+                    types.LiveClientRealtimeInput(
+                        media_chunks=[
+                            types.Blob(data=f.data.tobytes(), mime_type="audio/pcm")
+                        ]
+                    )
+                )
+        else:
+            self._queue_msg(
+                types.LiveClientRealtimeInput(
+                    media_chunks=[
+                        types.Blob(data=frame.data.tobytes(), mime_type="audio/pcm")
+                    ]
+                )
             )
-        )
 
     def _queue_msg(self, msg: ClientEvents) -> None:
         if not self._closed: