Merge pull request #13 from jquagga/faster_whisper

Add a faster_whisper function

docker-compose.yml

@@ -5,9 +5,11 @@ services:
       hostname: ttt
       restart: unless-stopped
       volumes:
-        - ../trunk-recorder-config/media/transcribe:/app/media/transcribe
+        - ./media/transcribe:/app/media/transcribe
         - ./destinations.csv:/app/destinations.csv:ro
+        - ./models:/app/models #faster_whisper models
       user: "1000"
       environment:
+        - TZ=America/New_York
         - TTT_WHISPERCPP_URL=${TTT_WHISPERCPP_URL}
-        - TTT_DEEPGRAM_KEY=${TTT_DEEPGRAM_KEY}
+        - TTT_DEEPGRAM_KEY=${TTT_DEEPGRAM_KEY}

ttt.py

@@ -33,11 +33,15 @@ def transcribe_call(destinations):
         calljson = json.loads(calljson)
 
         # Send the json and audiofile to a function to transcribe
-        # If TTT_DEEPGRAM_KEY is set, use deepgram, else whispercpp
-        if deepgram_key := os.environ.get("TTT_DEEPGRAM_KEY", False):
+        # If TTT_DEEPGRAM_KEY is set, use deepgram, else
+        # if TTT_WHISPER_URL is set, use whisper.cpp else
+        # fasterwhisper deepgram_key := whispercpp_url :=
+        if os.environ.get("TTT_DEEPGRAM_KEY", False):
             calljson = transcribe_deepgram(calljson, audiofile)
-        else:
+        elif os.environ.get("TTT_WHISPERCPP_URL", False):
             calljson = transcribe_whispercpp(calljson, audiofile)
+        else:
+            calljson = transcribe_fasterwhisper(calljson, audiofile)
 
         # Ok, we have text back, send for notification
         send_notifications(calljson, destinations)
@@ -50,11 +54,6 @@ def transcribe_call(destinations):
 def transcribe_whispercpp(calljson, audiofile):
     whisper_url = os.environ.get("TTT_WHISPERCPP_URL", "http://whisper:8080")
 
-    # Check if we are running behind
-    queue_time = float(datetime.now().timestamp()) - calljson["start_time"]
-    if queue_time > 180:
-        print("Queue exceeds 3 minutes")
-
     # Now send the files over to whisper for transcribing
     files = {
         "file": (None, audiofile.read_bytes()),
@@ -77,6 +76,35 @@ def transcribe_whispercpp(calljson, audiofile):
     return calljson
 
 
+def transcribe_fasterwhisper(calljson, audiofile):
+    from faster_whisper import WhisperModel
+
+    model_size = os.environ.get("TTT_FASTERWHISPER_MODEL", "distil-large-v3")
+    # model_size = "distil-large-v3"
+
+    # Run on GPU with FP16
+    # model = WhisperModel(model_size, device="cuda",
+    # compute_type="float16", download_root="models")
+
+    # or run on GPU with INT8
+    # model = WhisperModel(model_size, device="cuda",
+    # compute_type="int8_float16", download_root="models")
+    # or run on CPU with INT8
+    model = WhisperModel(
+        model_size, device="cpu", compute_type="int8", download_root="models"
+    )
+
+    # This whisper wants the path, not bytes but we need to cast it from pathlib to str
+    audiofile = str(audiofile)
+    segments, info = model.transcribe(audiofile, beam_size=5, vad_filter=True)
+
+    calltext = "".join(segment.text for segment in segments)
+
+    calljson["text"] = calltext
+
+    return calljson
+
+
 def transcribe_deepgram(calljson, audiofile):
     deepgram_key = os.environ.get("TTT_DEEPGRAM_KEY")
     headers = {