pepperplus/pepperplus-cb
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/dev' into refactor/zmq-internal-socket-behaviour

Browse Source 
# Conflicts:
#	src/control_backend/agents/ri_command_agent.py
#	src/control_backend/agents/ri_communication_agent.py
#	src/control_backend/api/v1/endpoints/command.py
#	src/control_backend/main.py
#	test/integration/api/endpoints/test_command_endpoint.py
This commit is contained in:
Twirre Meulenbelt
2025-11-05 12:16:18 +01:00
parent b008562554 da30c392aa
commit a1918049e5
30 changed files with 443 additions and 211 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/control_backend/agents/transcription/speech_recognizer.py
View File

@@ -75,7 +75,8 @@ class MLXWhisperSpeechRecognizer(SpeechRecognizer):
self.model_name = "mlx-community/whisper-small.en-mlx"
def load_model(self):
if self.was_loaded: return
if self.was_loaded:
return
# There appears to be no dedicated mechanism to preload a model, but this `get_model` does
# store it in memory for later usage
ModelHolder.get_model(self.model_name, mx.float16)
@@ -83,9 +84,9 @@ class MLXWhisperSpeechRecognizer(SpeechRecognizer):
def recognize_speech(self, audio: np.ndarray) -> str:
self.load_model()
return mlx_whisper.transcribe(audio,
path_or_hf_repo=self.model_name,
decode_options=self._get_decode_options(audio))["text"]
return mlx_whisper.transcribe(
audio, path_or_hf_repo=self.model_name, decode_options=self._get_decode_options(audio)
)["text"]
return mlx_whisper.transcribe(audio, path_or_hf_repo=self.model_name)["text"].strip()
@@ -95,12 +96,13 @@ class OpenAIWhisperSpeechRecognizer(SpeechRecognizer):
self.model = None
def load_model(self):
if self.model is not None: return
if self.model is not None:
return
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
self.model = whisper.load_model("small.en", device=device)
def recognize_speech(self, audio: np.ndarray) -> str:
self.load_model()
return whisper.transcribe(self.model,
audio,
decode_options=self._get_decode_options(audio))["text"]
return whisper.transcribe(
self.model, audio, decode_options=self._get_decode_options(audio)
)["text"]