Merge pull request #837 from ftnext/feature/pass-prompt-openai

Author: ftnext

speech_recognition/recognizers/whisper_api/base.py

@@ -1,7 +1,10 @@
+import logging
 from io import BytesIO
 
 from speech_recognition.audio import AudioData
 
+logger = logging.getLogger(__name__)
+
 
 class OpenAICompatibleRecognizer:
     def __init__(self, client) -> None:
@@ -16,7 +19,10 @@ def recognize(self, audio_data: "AudioData", model: str, **kwargs) -> str:
         wav_data = BytesIO(audio_data.get_wav_data())
         wav_data.name = "SpeechRecognition_audio.wav"
 
+        parameters = {"model": model, **kwargs}
+        logger.debug(parameters)
+
         transcript = self.client.audio.transcriptions.create(
-            file=wav_data, model=model, **kwargs
+            file=wav_data, **parameters
         )
         return transcript.text

speech_recognition/recognizers/whisper_api/openai.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import logging
 from typing import Literal
 
 from typing_extensions import Unpack
@@ -65,16 +66,31 @@ def recognize(
     parser = argparse.ArgumentParser()
     parser.add_argument("audio_file")
     parser.add_argument(
-        "--model", choices=get_args(WhisperModel), default="whisper-1"
+        "-m", "--model", choices=get_args(WhisperModel), default="whisper-1"
     )
     parser.add_argument("-l", "--language")
+    parser.add_argument("-p", "--prompt")
+    parser.add_argument("-v", "--verbose", action="store_true")
     args = parser.parse_args()
 
+    if args.verbose:
+        speech_recognition_logger = logging.getLogger("speech_recognition")
+        speech_recognition_logger.setLevel(logging.DEBUG)
+
+        console_handler = logging.StreamHandler()
+        console_formatter = logging.Formatter(
+            "%(asctime)s | %(levelname)s | %(name)s:%(funcName)s:%(lineno)d - %(message)s"
+        )
+        console_handler.setFormatter(console_formatter)
+        speech_recognition_logger.addHandler(console_handler)
+
     audio_data = sr.AudioData.from_file(args.audio_file)
+
+    recognize_args = {"model": args.model}
     if args.language:
-        transcription = recognize(
-            None, audio_data, model=args.model, language=args.language
-        )
-    else:
-        transcription = recognize(None, audio_data, model=args.model)
+        recognize_args["language"] = args.language
+    if args.prompt:
+        recognize_args["prompt"] = args.prompt
+
+    transcription = recognize(None, audio_data, **recognize_args)
     print(transcription)