WhisperX to word-level tsv

from pathlib import Path

BASE = Path("/Users/joregan/Playing/hsi/audio/whisperx-json/")
TSVBASE = Path("/Users/joregan/Playing/hsi/audio/whisperx-tsv/")

import json

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

%pip install num2words

Requirement already satisfied: num2words in /Users/joregan/opt/anaconda3/envs/hf/lib/python3.9/site-packages (0.5.13)
Requirement already satisfied: docopt>=0.6.2 in /Users/joregan/opt/anaconda3/envs/hf/lib/python3.9/site-packages (from num2words) (0.6.2)
Note: you may need to restart the kernel to use updated packages.

from num2words import num2words

import re

def get_words(word):
    if re.match("^[2-9]0s$", word):
        oword = num2words(int(word[0:-1]), lang="en")
        return oword[:-1] + "ies"
    if re.match("^[0-9]+(?:th|st|nd|rd)$", word):
        return num2words(int(word[0:-2]), lang="en", to="ordinal")
    if re.match("^[0-9]+(?:,[0-9][0-9][0-9])+$", word):
        return num2words(int(word.replace(",", "")), lang="en")
    if re.match("^\$[0-9]+(?:,[0-9][0-9][0-9])*$", word):
        return num2words(int(word[1:].replace(",", "")), lang="en") + " dollars"
    if re.match("^20[1-9][0-9]$", word):
        return "twenty " + num2words(int(word[2:]), lang="en")
    if re.match("^[0-9]+$", word):
        return num2words(int(word), lang="en")
    return word

for jsonfile in BASE.glob("*.json"):
    last_end = 0.0
    outtsv = TSVBASE / f"{jsonfile.stem}.tsv"
    with open(str(jsonfile)) as inf, open(str(outtsv), "w") as outf:
        data = json.load(inf)
        for segment in data['segments']:
            for word in segment['words']:
                if not "start" in word:
                    start = last_end
                    end = last_end + 0.05
                    text = get_words(word['word'])
                else:
                    start = word['start']
                    end = word['end']
                    text = word['word']
                outf.write(str(start) + "\t" + str(end) + "\t" + text + "\n")
                last_end = end

TSVBASE = Path("/Users/joregan/Playing/hsi/audio/whisperx-tsv-sentence/")

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

def get_words_punct(word):
    m = re.match("^([^A-Za-z0-9]+)(.*[0-9].*)([^A-Za-z0-9]+)$", word)
    if m:
        pre = m.group(1)
        mid = m.group(2)
        post = m.group(3)
        return pre + get_words(mid) + post
    m = re.match("^([^A-Za-z0-9]+)(.*[0-9][A-Za-z0-9]+)$", word)
    if m:
        pre = m.group(1)
        mid = m.group(2)
        return pre + get_words(mid)
    m = re.match("^([A-Za-z0-9].*[0-9].*)([^A-Za-z0-9]+)$", word)
    if m:
        mid = m.group(1)
        post = m.group(2)
        return get_words(mid) + post
    return get_words(word)

for jsonfile in BASE.glob("*.json"):
    outtsv = TSVBASE / f"{jsonfile.stem}.tsv"
    with open(str(jsonfile)) as inf, open(str(outtsv), "w") as outf:
        data = json.load(inf)
        for segment in data['segments']:
            start = str(segment['start'])
            end = str(segment['end'])
            text = segment['text'].strip()
            normed = [get_words_punct(x) for x in text.split(" ")]
            outf.write(start + "\t" + end + "\t" + " ".join(normed) + "\n")