Extract segments for ctc-forced-aligner

Built on earlier notebook, using data extracted with this notebook

from pathlib import Path

segments = {}
count = 0
for tsvfile in Path("/tmp/textgrid_cut").glob("*.csv"):
    current = []
    filestem = tsvfile.stem
    with open(str(tsvfile)) as tsvf:
        for line in tsvf.readlines():
            line = line.strip()
            if line == "":
                continue
            parts = line.split("\t")
            if len(parts) != 3:
                print(line)
                continue
            start = float(parts[0])
            end = float(parts[1])
            text = parts[2]
            duration = (end - start)
            current.append((start, end, text))
        segments[filestem] = current

237.16626686591627	237.93940299427976

len(segments)

70

from pydub import AudioSegment

parameters = ["-ac", "1", "-acodec", "pcm_s16le", "-ar", "16000"]
matcha_data = {}

outdir = Path("/tmp/audio_clips")

if not outdir.is_dir():
    outdir.mkdir()

for scene in segments:
    parts = scene.split("_")
    if parts[-1] == "main":
        speaker = int(parts[1]) - 2
    else:
        if parts[1] == "3":
            speaker = 5
        else:
            speaker = 6
    if speaker not in matcha_data:
        matcha_data[speaker] = []
    
    orig_audio = AudioSegment.from_file(f"/Users/joregan/Playing/hsi/audio/audio_cut/{scene}.wav")

    for seg in segments[scene]:
        start = int(seg[0] * 1000)
        end = int(seg[1] * 1000)
        text = seg[2]

        chunk = orig_audio[start:end]
        chunk_basename = f"{scene}_{start}_{end}.wav"
        chunk_name = str(outdir / chunk_basename)

        chunk.export(chunk_name, format="wav", parameters=parameters)
        with open(str(outdir / chunk_basename.replace(".wav", ".txt")), "w") as txtout:
            txtout.write(text)

Aligner built with this docker

docker run -it --entrypoint /bin/bash -v'/home/joregan/audio_clips:/audio' --gpus all ctc-forced-aligner

for i in /audio/*.wav;do ctc-forced-aligner --audio_path $i --text_path $(echo $i|sed -e 's/wav/txt/') --language en ;done