Skip to content

armgit / macro

Public Imported from GitHub
0

Macro is a unified workspace for teams: email, chat, docs, tasks, agents, calls, and CRM — @-linked together with shared AI memory.

feat(calls): support multiple languages for call transcriptions (#6093)

Will Hutchinson8d8f05d4c42d
diff --git a/services/transcription/requirements.txt b/services/transcription/requirements.txt
index bf9d28c183..d98ef6ba34 100644
--- a/services/transcription/requirements.txt
+++ b/services/transcription/requirements.txt
@@ -2,3 +2,4 @@ python-dotenv
livekit-agents[silero]==1.5.7
httpx
resemblyzer==0.1.4
+setuptools==80.10.2 # resemblyzer's webrtcvad dependency imports pkg_resources
diff --git a/services/transcription/test_transcriber.py b/services/transcription/test_transcriber.py
new file mode 100644
index 0000000000..60a775e28f
--- /dev/null
+++ b/services/transcription/test_transcriber.py
@@ -0,0 +1,36 @@
+import unittest
+from unittest.mock import patch, sentinel
+
+import transcriber
+
+
+class TranscriberConfigurationTest(unittest.TestCase):
+ def test_configures_multilingual_deepgram_transcription(self):
+ with (
+ patch.object(
+ transcriber.inference,
+ "STT",
+ return_value=sentinel.stt,
+ ) as create_stt,
+ patch.object(transcriber.Agent, "__init__", return_value=None) as init_agent,
+ ):
+ transcriber.Transcriber(
+ participant_identity="participant-id",
+ channel_id="channel-id",
+ http_client=sentinel.http_client,
+ )
+
+ create_stt.assert_called_once()
+ args, kwargs = create_stt.call_args
+ self.assertEqual(args, ("deepgram/nova-3",))
+ self.assertEqual(kwargs["language"], "multi")
+ self.assertTrue(kwargs["extra_kwargs"]["diarize"])
+ self.assertNotIn("detect_language", kwargs["extra_kwargs"])
+ init_agent.assert_called_once_with(
+ instructions="Transcribe user speech.",
+ stt=sentinel.stt,
+ )
+
+
+if __name__ == "__main__":
+ unittest.main()
diff --git a/services/transcription/transcriber.py b/services/transcription/transcriber.py
index 928ab6a1d7..5e7a862302 100644
--- a/services/transcription/transcriber.py
+++ b/services/transcription/transcriber.py
@@ -161,7 +161,7 @@ class Transcriber(Agent):
instructions="Transcribe user speech.",
stt=inference.STT(
"deepgram/nova-3",
- language="en-US",
+ language="multi",
extra_kwargs={
# ms of silence before Deepgram finalizes an utterance.
# Library default is 25ms, which cuts hesitant speakers mid-thought.