Add speech_only.py script

scripts/speech_only.py

@@ -0,0 +1,60 @@
+import sys
+from silero_vad import load_silero_vad
+from silero_vad import read_audio , get_speech_timestamps , save_audio
+from silero_vad import VADIterator , collect_chunks
+import torch
+import json
+import os
+import argparse
+import logging
+from pathlib import Path
+
+logging.basicConfig(
+    format="%(asctime)s | %(levelname)s | %(name)s | %(message)s",
+    datefmt="%Y-%m-%d %H:%M:%S",
+    level=os.environ.get("LOG_LEVEL", "INFO").upper())
+LOGGER = logging.getLogger(__name__)
+
+def main(args):
+    parser = argparse.ArgumentParser(args)
+    parser.add_argument("--folder", type=str, required=True, help="folder" )
+    parser.add_argument("--sfx", type=str, required=False, default="wav",help="audio suffix" )
+    parser.add_argument("--sr", type=int, required=False, default="16000",help="sampling rate" )
+    parser.add_argument("--out_folder", type=str, required=False, default=None,help="output folder" )
+    parser.add_argument("--out_file", type=str, required=True, help="json output" )
+    parser.add_argument("--reverse", action="store_true", help="reverse processing order in folder")
+    args = parser.parse_args()
+
+    model = load_silero_vad(onnx=True)
+    sr=args.sr
+    out_folder=args.out_folder
+    if out_folder:
+        Path(out_folder).mkdir(parents=True, exist_ok=True)
+        save_wav=True
+    audio_list = sorted(os.listdir(args.folder),reverse=args.reverse)
+    with open(args.out_file, "w") as outfile:
+        for audiofile in audio_list:
+            if audiofile.endswith(args.sfx):
+                audiopath = os.path.join(args.folder, audiofile)
+                audiofile = audiofile.replace(args.sfx,"wav")
+                audio_id  = Path(audiofile).stem
+                out_audio=f"{out_folder}/{audiofile}"
+                if os.path.isfile(out_audio):
+                    LOGGER.info(f"skipping {audiopath}")
+                    continue
+                LOGGER.info(f"processing {audiopath}")
+                wav = read_audio(audiopath, sampling_rate=sr)
+                speech_timestamps = get_speech_timestamps(wav, model, sampling_rate=sr)
+
+                if speech_timestamps:
+                    LOGGER.info(f"vad-processed {audiofile}: {len(speech_timestamps)} chunks")
+                    for turn in speech_timestamps:
+                        entry = {"audio_id": audio_id, "offset": turn['start']/sr, "duration":  (turn['end']-turn['start'])/sr}
+                        json.dump(entry, outfile)
+                        outfile.write('\n')
+                    if save_wav:
+                        save_audio(out_audio,collect_chunks(speech_timestamps, wav), sampling_rate=sr)
+                        LOGGER.info(f"vad-processed {audiofile} written to {out_audio}")
+
+if __name__ == '__main__':
+    main(sys.argv)