@webship.co/workspace 1.0.0-beta1 → 1.0.0-beta2

package/core/scripts/libs/elevenlabs.py

@@ -0,0 +1,627 @@
+#!/usr/bin/env python3
+"""ElevenLabs for the final voiceover of a narrated video. Standard library only.
+
+Run through videos/cmd-tools-elevenlabs.sh, which reads the key from settings.yml and hands it over in
+ELEVENLABS_API_KEY. Anything that spends credits prints a plan and stops unless --confirm is given.
+Exit codes: 0 ok, 1 error, 2 usage, 3 no key, 4 plan only, 5 sections do not fit their windows.
+"""
+
+import argparse
+import datetime
+import hashlib
+import json
+import os
+import re
+import shutil
+import subprocess
+import sys
+import urllib.error
+import urllib.parse
+import urllib.request
+import uuid
+
+OK, ERROR, USAGE, NO_KEY, PLAN_ONLY, NO_FIT = 0, 1, 2, 3, 4, 5
+DEFAULT_BASE = 'https://api.elevenlabs.io'
+DEFAULT_MODEL = 'eleven_v4'
+DEFAULT_FORMAT = 'mp3_44100_128'
+SPEECH_CHARS_PER_SECOND = 15.0
+SLACK = 0.05
+
+HEADING = re.compile(
+    r'^##\s+(\S+)\s+(.+?)\s+\(((?:\d+:)?\d{1,2}:\d{2}(?:\.\d+)?)\s+to\s+((?:\d+:)?\d{1,2}:\d{2}(?:\.\d+)?)\)\s*$')
+HINTS = {
+    401: 'ElevenLabs rejected the key.',
+    403: 'The key is not allowed to do that.',
+    404: 'Not found (check the voice id).',
+    429: 'ElevenLabs is rate limiting this key. Wait, then run it again; finished sections are kept.',
+}
+
+
+class Stop(Exception):
+    def __init__(self, message, code=ERROR):
+        super().__init__(message)
+        self.code = code
+
+
+def emit(args, data, lines):
+    print(json.dumps(data, indent=2) if args.json else '\n'.join(lines))
+
+
+def mmss(seconds):
+    return '%02d:%04.1f' % (int(seconds // 60), seconds % 60)
+
+
+def seconds_of(stamp):
+    total = 0.0
+    for part in stamp.split(':'):
+        total = total * 60 + float(part)
+    return total
+
+
+def sha(*parts):
+    return hashlib.sha256(json.dumps(parts, sort_keys=True).encode()).hexdigest()
+
+
+def stem_of(path):
+    return os.path.splitext(os.path.basename(path))[0]
+
+
+def safe_id(text):
+    return re.sub(r'[^A-Za-z0-9_-]', '_', text)
+
+
+def read_bytes(path):
+    with open(path, 'rb') as handle:
+        return handle.read()
+
+
+def read_json(path):
+    with open(path, encoding='utf-8') as handle:
+        return json.load(handle)
+
+
+def ensure_free(paths, force):
+    for path in paths:
+        if os.path.exists(path) and not force:
+            raise Stop('%s already exists. Pass --force to replace it. Nothing was sent.' % path)
+
+
+def track_of(out):
+    return os.path.splitext(out)[0] + '-voiceover.mp3'
+
+
+def write_atomic(path, data, force):
+    if os.path.exists(path) and not force:
+        raise Stop('%s already exists. Pass --force to replace it.' % path)
+    folder = os.path.dirname(os.path.abspath(path))
+    os.makedirs(folder, exist_ok=True)
+    part = os.path.join(folder, '.%s.%s.part' % (os.path.basename(path), uuid.uuid4().hex[:8]))
+    with open(part, 'wb') as handle:
+        handle.write(data)
+    os.replace(part, path)
+
+
+# ---- the API
+
+def key():
+    value = os.environ.get('ELEVENLABS_API_KEY', '').strip()
+    if not value:
+        raise Stop('No ElevenLabs key is set. Add it at /settings (ElevenLabs), or run: bash cmd-tools-elevenlabs.sh -a init', NO_KEY)
+    if not re.fullmatch(r'[A-Za-z0-9_.-]+', value):
+        raise Stop('The ElevenLabs key has characters a key never has. Check the value in settings.yml.', NO_KEY)
+    return value
+
+
+def call(method, path, body=None, content_type=None, query=None):
+    url = os.environ.get('ELEVENLABS_API_BASE', DEFAULT_BASE).rstrip('/') + path
+    if query:
+        url += '?' + urllib.parse.urlencode(query)
+    headers = {'xi-api-key': key()}
+    if content_type:
+        headers['Content-Type'] = content_type
+    request = urllib.request.Request(url, data=body, method=method, headers=headers)
+    try:
+        with urllib.request.urlopen(request, timeout=int(os.environ.get('ELEVENLABS_TIMEOUT', '300'))) as response:
+            return response.status, response.headers, response.read()
+    except urllib.error.HTTPError as error:
+        return error.code, error.headers, error.read()
+    except (urllib.error.URLError, TimeoutError, OSError) as error:
+        raise Stop('Could not reach ElevenLabs (%s).' % getattr(error, 'reason', error))
+
+
+ID_LIKE = re.compile(r'[A-Za-z0-9]{16,32}')
+
+
+def resolve_voice(value):
+    """A voice given by name (Brian) becomes its id on this account; an id is left alone."""
+    if ID_LIKE.fullmatch(value):
+        return value
+    status, _, body = call('GET', '/v2/voices', query={'search': value, 'page_size': '30'})
+    if status != 200:
+        raise Stop('Could not look up the voice "%s". %s' % (value, explain(status, body)))
+    voices = json.loads(body).get('voices', [])
+    want = value.strip().lower()
+    hit = (next((v for v in voices if (v.get('name') or '').lower() == want), None)
+           or next((v for v in voices if (v.get('name') or '').lower().startswith(want)), None))
+    if not hit:
+        raise Stop('No voice named "%s" on this account. List them with -a voices.' % value, USAGE)
+    return hit['voice_id']
+
+
+def explain(status, body):
+    detail = ''
+    try:
+        found = json.loads(body.decode('utf-8', 'replace')).get('detail')
+        if isinstance(found, dict):
+            detail = found.get('message') or found.get('status') or ''
+        elif isinstance(found, list):
+            detail = '; '.join(str(item.get('msg', item)) if isinstance(item, dict) else str(item) for item in found[:3])
+        elif found:
+            detail = str(found)
+    except Exception:
+        detail = body.decode('utf-8', 'replace')[:160].replace('\n', ' ')
+    hint = HINTS.get(status) or ('ElevenLabs had a server error; run it again later.' if status >= 500 else '')
+    return ' '.join(part for part in ('HTTP %d.' % status, hint, detail) if part)
+
+
+def multipart(fields, files):
+    boundary = uuid.uuid4().hex
+    parts = []
+    for name, value in fields.items():
+        parts.append(('--%s\r\nContent-Disposition: form-data; name="%s"\r\n\r\n%s\r\n' % (boundary, name, value)).encode())
+    for name, filename, data in files:
+        parts.append(('--%s\r\nContent-Disposition: form-data; name="%s"; filename="%s"\r\n'
+                      'Content-Type: application/octet-stream\r\n\r\n' % (boundary, name, filename)).encode() + data + b'\r\n')
+    parts.append(('--%s--\r\n' % boundary).encode())
+    return b''.join(parts), 'multipart/form-data; boundary=%s' % boundary
+
+
+def account():
+    status, _, body = call('GET', '/v1/user/subscription')
+    if status != 200:
+        return None, explain(status, body)
+    found = json.loads(body)
+    used, limit = found.get('character_count'), found.get('character_limit')
+    return {
+        'tier': found.get('tier'), 'status': found.get('status'), 'used': used, 'limit': limit,
+        'remaining': limit - used if isinstance(used, int) and isinstance(limit, int) else None,
+        'resets': found.get('next_character_count_reset_unix'),
+    }, ''
+
+
+def remaining_characters():
+    try:
+        found, _ = account()
+    except Stop:
+        return None
+    return found['remaining'] if found else None
+
+
+# Brand names the voices misread: a client heard "Var-dough" for Webship in a final video. The respelling is
+# applied ONLY to the text sent to the voice. Scripts, captions, transcripts, character counts and the
+# saved section text keep the real spelling. ELEVENLABS_PRONUNCIATION ("Word=Respelling; Word=Respelling",
+# from elevenlabs.pronunciation in settings.yml) adds to or overrides these.
+PRONUNCIATIONS = {'Webship': 'Var-dot', 'Webship': 'Var-base'}
+
+
+def pronunciations():
+    table = dict(PRONUNCIATIONS)
+    for pair in os.environ.get('ELEVENLABS_PRONUNCIATION', '').split(';'):
+        word, _, said = pair.partition('=')
+        if word.strip() and said.strip():
+            table[word.strip()] = said.strip()
+    return table
+
+
+def respell(text):
+    for word, said in pronunciations().items():
+        text = re.sub(r'(?<![\w-])%s(?![\w-])' % re.escape(word), said, text)
+    return text
+
+
+def synthesize(text, voice, model, fmt):
+    body = json.dumps({'text': respell(text), 'model_id': model}).encode()
+    status, headers, data = call('POST', '/v1/text-to-speech/' + urllib.parse.quote(voice, safe=''), body=body,
+                                 content_type='application/json', query={'output_format': fmt})
+    if status != 200 or not data or 'json' in (headers.get('content-type') or ''):
+        raise Stop('Speech failed. ' + explain(status, data))
+    return data, {'request_id': headers.get('request-id'), 'characters': headers.get('x-character-count')}
+
+
+# ---- media
+
+def need_tools():
+    for tool in ('ffmpeg', 'ffprobe'):
+        if not shutil.which(tool):
+            raise Stop('%s is needed and was not found on PATH.' % tool)
+
+
+def run(command):
+    done = subprocess.run(command, capture_output=True, text=True)
+    if done.returncode != 0:
+        raise Stop('%s failed: %s' % (command[0], (done.stderr or '').strip().splitlines()[-1:] or 'no output'))
+
+
+def duration(path):
+    try:
+        done = subprocess.run(['ffprobe', '-v', 'error', '-show_entries', 'format=duration', '-of', 'csv=p=0', path],
+                              capture_output=True, text=True, timeout=60)
+        return float(done.stdout.strip())
+    except Exception:
+        return None
+
+
+def mux(video, mix_wav, out, length, force):
+    ensure_free([out, track_of(out)], force)
+    part = os.path.join(os.path.dirname(os.path.abspath(out)), '.%s.part' % os.path.basename(out))
+    run(['ffmpeg', '-y', '-loglevel', 'error', '-i', video, '-i', mix_wav, '-map', '0:v:0', '-map', '1:a:0',
+         '-c:v', 'copy', '-c:a', 'aac', '-b:a', '192k', '-movflags', '+faststart', '-t', '%.3f' % length, '-f', 'mp4', part])
+    os.replace(part, out)
+    track = track_of(out)
+    run(['ffmpeg', '-y', '-loglevel', 'error', '-i', mix_wav, '-c:a', 'libmp3lame', '-b:a', '192k', track])
+    poster = os.path.splitext(video)[0] + '-poster.jpg'
+    new_poster = os.path.splitext(out)[0] + '-poster.jpg'
+    if os.path.exists(poster) and (force or not os.path.exists(new_poster)):
+        shutil.copyfile(poster, new_poster)
+    os.remove(mix_wav)
+    return track, new_poster if os.path.exists(new_poster) else None
+
+
+# ---- the script
+
+def clean(line):
+    text = re.sub(r'\*\(optional\)\*', '', line)
+    text = re.sub(r'\[([^\]]+)\]\([^)]*\)', r'\1', text)
+    return ' '.join(re.sub(r'[*`]', '', text).split())
+
+
+def load_sections(path, drop_optional=False):
+    try:
+        with open(path, encoding='utf-8') as handle:
+            lines = handle.read().splitlines()
+    except OSError as error:
+        raise Stop('Cannot read the script: %s' % error)
+    sections, current = [], None
+    for line in lines:
+        found = HEADING.match(line)
+        if found:
+            current = {'id': found.group(1), 'title': found.group(2), 'start': seconds_of(found.group(3)),
+                       'end': seconds_of(found.group(4)), 'spoken': [], 'optional': []}
+            sections.append(current)
+        elif line.startswith('## '):
+            current = None
+        elif current is not None and line.startswith('>'):
+            spoken = clean(line.lstrip('>'))
+            if spoken:
+                (current['optional'] if '*(optional)*' in line else current['spoken']).append((len(current['spoken']) + len(current['optional']), spoken))
+    if not sections:
+        raise Stop('No timed sections found. Each section needs a heading like "## D01 Opening (00:00.0 to 00:21.4)" and its '
+                   'spoken lines as "> " quotes. See the elevenlabs-voiceover skill.')
+    result, previous_end = [], 0.0
+    for section in sections:
+        if section['end'] <= section['start']:
+            raise Stop('Section %s ends before it starts.' % section['id'])
+        if section['start'] < previous_end - SLACK:
+            raise Stop('Section %s starts before the previous section ends.' % section['id'])
+        previous_end = section['end']
+        kept = sorted(section['spoken'] + ([] if drop_optional else section['optional']))
+        text = ' '.join(item[1] for item in kept)
+        if not text:
+            continue
+        result.append({
+            'id': section['id'], 'title': section['title'], 'start': section['start'], 'end': section['end'],
+            'window': round(section['end'] - section['start'], 3), 'text': text, 'chars': len(text),
+            'optional_chars': sum(len(item[1]) for item in section['optional']),
+        })
+    if not result:
+        raise Stop('The script has sections but no spoken lines ("> " quotes).')
+    return result
+
+
+def check_inputs(args):
+    for attribute in ('script', 'video', 'audio'):
+        value = getattr(args, attribute, None)
+        if value and not os.path.isfile(value):
+            raise Stop('%s not found: %s' % (attribute.capitalize(), value), USAGE)
+
+
+def work_dir(args):
+    return args.work or os.path.join(os.path.dirname(os.path.abspath(args.out)), stem_of(args.out) + '-voiceover-sections')
+
+
+def audio_ext(fmt):
+    if not re.fullmatch(r'(mp3|wav)_\d+(_\d+)?', fmt):
+        raise Stop('The format must be an mp3_* or wav_* one, for example %s.' % DEFAULT_FORMAT, USAGE)
+    return fmt.split('_')[0]
+
+
+# ---- commands
+
+def cmd_sections(args):
+    check_inputs(args)
+    sections = load_sections(args.script, args.drop_optional)
+    lines = ['%-5s %-34s %-18s %7s %6s %9s' % ('id', 'section', 'window', 'seconds', 'chars', 'optional')]
+    for s in sections:
+        lines.append('%-5s %-34s %-18s %7.1f %6d %9d' % (s['id'], s['title'][:34], '%s to %s' % (mmss(s['start']), mmss(s['end'])),
+                                                          s['window'], s['chars'], s['optional_chars']))
+    lines.append('%d sections, %d characters.' % (len(sections), sum(s['chars'] for s in sections)))
+    emit(args, {'sections': sections, 'characters': sum(s['chars'] for s in sections)}, lines)
+    return OK
+
+
+def cmd_voices(args):
+    query = {'page_size': str(args.limit)}
+    if args.search:
+        query['search'] = args.search
+    status, _, body = call('GET', '/v2/voices', query=query)
+    if status != 200:
+        raise Stop('Could not list voices. ' + explain(status, body))
+    voices = json.loads(body).get('voices', [])
+    rows = [{'id': v.get('voice_id'), 'name': v.get('name'), 'category': v.get('category'), 'labels': v.get('labels') or {}} for v in voices]
+    lines = ['%-24s %-22s %-10s %s' % ('id', 'name', 'category', 'labels')]
+    for r in rows:
+        lines.append('%-24s %-22s %-10s %s' % (r['id'], (r['name'] or '')[:22], r['category'] or '', ', '.join('%s=%s' % kv for kv in r['labels'].items())))
+    emit(args, {'voices': rows}, lines)
+    return OK
+
+
+def cmd_account(args):
+    found, why = account()
+    if not found:
+        raise Stop('The key works for requests but the account cannot be read. %s Grant the key "User: read" to see credits.' % why)
+    resets = datetime.datetime.fromtimestamp(found['resets'], datetime.timezone.utc).strftime('%Y-%m-%d') if found['resets'] else 'unknown'
+    emit(args, found, ['Plan: %s (%s)' % (found['tier'], found['status']), 'Characters used: %s of %s' % (found['used'], found['limit']),
+                       'Remaining: %s' % found['remaining'], 'Resets: %s' % resets])
+    return OK
+
+
+def cmd_speak(args):
+    try:
+        with open(args.text_file, encoding='utf-8') as handle:
+            text = handle.read().strip()
+    except OSError:
+        text = ''
+    if not text:
+        raise Stop('The text file is missing or empty: %s' % args.text_file, USAGE)
+    key()
+    remaining = remaining_characters()
+    plan = {'voice': args.voice, 'model': args.model, 'format': args.format, 'characters': len(text), 'remaining': remaining, 'out': args.out}
+    if not args.confirm:
+        emit(args, dict(plan, confirmed=False), ['Plan only, nothing was sent.', 'Voice %s, model %s, %d characters.' % (args.voice, args.model, len(text)),
+                                                  'Credits left: %s.' % ('unknown' if remaining is None else remaining), 'Run again with --confirm to create %s.' % args.out])
+        return PLAN_ONLY
+    ensure_free([args.out], args.force)
+    data, meta = synthesize(text, args.voice, args.model, args.format)
+    write_atomic(args.out, data, args.force)
+    emit(args, dict(plan, confirmed=True, bytes=len(data), seconds=duration(args.out), **meta), ['Wrote %s (%d bytes).' % (args.out, len(data))])
+    return OK
+
+
+def cmd_isolate(args):
+    check_inputs(args)
+    key()
+    seconds = duration(args.audio)
+    plan = {'audio': args.audio, 'seconds': seconds, 'out': args.out}
+    if not args.confirm:
+        emit(args, dict(plan, confirmed=False), ['Plan only, nothing was sent.', 'Would clean %s (%s seconds) and write %s.' % (args.audio, seconds, args.out),
+                                                  'Run again with --confirm.'])
+        return PLAN_ONLY
+    ensure_free([args.out], args.force)
+    body, content_type = multipart({}, [('audio', os.path.basename(args.audio), read_bytes(args.audio))])
+    status, _, data = call('POST', '/v1/audio-isolation', body=body, content_type=content_type)
+    if status != 200 or not data:
+        raise Stop('Isolation failed. ' + explain(status, data))
+    write_atomic(args.out, data, args.force)
+    emit(args, dict(plan, confirmed=True, bytes=len(data)), ['Wrote %s (%d bytes).' % (args.out, len(data))])
+    return OK
+
+
+def cmd_voiceover(args):
+    check_inputs(args)
+    need_tools()
+    key()
+    ext = audio_ext(args.format)
+    sections = load_sections(args.script, args.drop_optional)
+    length = duration(args.video)
+    if not length:
+        raise Stop('Cannot read the video duration: %s' % args.video)
+    if sections[-1]['end'] > length + 1.0:
+        raise Stop('The script runs to %s but the video is only %s long. Is this the right draft?' % (mmss(sections[-1]['end']), mmss(length)))
+    work = work_dir(args)
+    for s in sections:
+        s['fingerprint'] = sha(s['text'], args.voice, args.model, args.format, pronunciations())
+        s['file'] = os.path.join(work, '%s.%s' % (safe_id(s['id']), ext))
+        meta_path = os.path.join(work, '%s.json' % safe_id(s['id']))
+        try:
+            s['cached'] = os.path.isfile(s['file']) and not args.redo and read_json(meta_path)['fingerprint'] == s['fingerprint']
+        except (OSError, ValueError, KeyError):
+            s['cached'] = False
+    todo = [s for s in sections if not s['cached']]
+    to_send = sum(s['chars'] for s in todo)
+    remaining = remaining_characters()
+    rows = [dict(id=s['id'], title=s['title'], window=s['window'], chars=s['chars'], rough_seconds=round(s['chars'] / SPEECH_CHARS_PER_SECOND, 1),
+                 may_be_long=s['chars'] / SPEECH_CHARS_PER_SECOND > s['window'] * 1.05, cached=s['cached'], has_optional=s['optional_chars'] > 0)
+            for s in sections]
+    plan = {'script': args.script, 'video': args.video, 'out': args.out, 'voice': args.voice, 'model': args.model, 'sections': rows,
+            'characters_to_send': to_send, 'remaining': remaining, 'enough': None if remaining is None else remaining >= to_send, 'work': work}
+    if not args.confirm:
+        lines = ['Plan only, nothing was sent.', 'Voice %s, model %s. Video %s long.' % (args.voice, args.model, mmss(length)),
+                 '%-5s %-30s %7s %6s %9s  %s' % ('id', 'section', 'window', 'chars', '~seconds', 'note')]
+        for r in rows:
+            note = 'already made, no cost' if r['cached'] else ('may run long' + (' (has an optional sentence)' if r['has_optional'] else '') if r['may_be_long'] else '')
+            lines.append('%-5s %-30s %6.1fs %6d %9.1f  %s' % (r['id'], r['title'][:30], r['window'], r['chars'], r['rough_seconds'], note))
+        lines += ['%d characters would be sent; credits left: %s.' % (to_send, 'unknown' if remaining is None else remaining),
+                  '"~seconds" is a rough estimate from the character count. Run again with --confirm to create the voiceover.']
+        emit(args, dict(plan, confirmed=False), lines)
+        return PLAN_ONLY
+    if remaining is not None and remaining < to_send:
+        raise Stop('Not enough credits: %d characters to send, %d left. Nothing was sent.' % (to_send, remaining))
+    ensure_free([args.out, track_of(args.out)], args.force)
+    os.makedirs(work, exist_ok=True)
+    for s in todo:
+        data, meta = synthesize(s['text'], args.voice, args.model, args.format)
+        write_atomic(s['file'], data, True)
+        with open(os.path.join(work, '%s.json' % safe_id(s['id'])), 'w') as handle:
+            json.dump(dict(meta, fingerprint=s['fingerprint'], id=s['id']), handle)
+    report = []
+    for s in sections:
+        seconds = duration(s['file'])
+        row = {'id': s['id'], 'title': s['title'], 'window': s['window'], 'audio_seconds': seconds, 'fit': 'ok', 'speedup': 1.0, 'audio': s['file']}
+        if seconds is None:
+            raise Stop('Cannot read the audio for section %s.' % s['id'])
+        if seconds > s['window'] + SLACK:
+            ratio = seconds / s['window']
+            if ratio <= args.max_speedup:
+                fitted = os.path.join(work, '%s-fit.wav' % safe_id(s['id']))
+                run(['ffmpeg', '-y', '-loglevel', 'error', '-i', s['file'], '-filter:a', 'atempo=%.5f' % ratio, '-ar', '44100', '-ac', '1', fitted])
+                row.update(fit='sped_up', speedup=round(ratio, 4), audio=fitted, audio_seconds=duration(fitted))
+            else:
+                row.update(fit='too_long', speedup=round(ratio, 4), over_by=round(seconds - s['window'], 2), has_optional=s['optional_chars'] > 0)
+        report.append(row)
+    too_long = [r for r in report if r['fit'] == 'too_long']
+    if too_long:
+        lines = ['Created %d new section(s), but %d %s not fit its window, so no video was made. Finished sections are kept.' % (len(todo), len(too_long), 'does' if len(too_long) == 1 else 'do')]
+        lines += ['  %s %s: %.1fs of speech in a %.1fs window (%.1fs over)%s' % (r['id'], r['title'], r['audio_seconds'], r['window'], r['over_by'],
+                                                                              ', has an optional sentence to drop' if r['has_optional'] else '') for r in too_long]
+        lines.append('Shorten the text in the script (or use --drop-optional), then run again: only the changed sections are sent.')
+        emit(args, {'confirmed': True, 'fit': False, 'sections': report}, lines)
+        return NO_FIT
+    mix = os.path.join(work, '_mix.wav')
+    chains = ['[%d:a]aformat=sample_rates=44100:channel_layouts=mono,adelay=%d:all=1[a%d]' % (i + 1, round(s['start'] * 1000), i) for i, s in enumerate(sections)]
+    chains.append('%samix=inputs=%d:duration=longest:normalize=0,apad=whole_dur=%.3f,atrim=end=%.3f[aout]'
+                  % (''.join('[a%d]' % i for i in range(len(sections))), len(sections), length, length))
+    run(['ffmpeg', '-y', '-loglevel', 'error', '-i', args.video] + [x for r in report for x in ('-i', r['audio'])] +
+        ['-filter_complex', ';'.join(chains), '-map', '[aout]', '-ar', '44100', '-ac', '2', mix])
+    track, poster = mux(args.video, mix, args.out, length, args.force)
+    summary = {'created_utc': datetime.datetime.now(datetime.timezone.utc).strftime('%Y-%m-%dT%H:%M:%SZ'), 'script': args.script, 'video': args.video,
+               'voice': args.voice, 'model': args.model, 'format': args.format, 'characters_sent': to_send, 'sections': report}
+    with open(os.path.splitext(args.out)[0] + '-voiceover.json', 'w') as handle:
+        json.dump(summary, handle, indent=2)
+    emit(args, dict(summary, confirmed=True, fit=True, out=args.out, track=track, poster=poster),
+         ['Wrote %s' % args.out, 'Voiceover track: %s' % track] + (['Poster: %s' % poster] if poster else []) +
+         ['%d sections placed, %d sped up (never more than %.0f%%), %d characters sent.' % (len(report), sum(r['fit'] == 'sped_up' for r in report),
+                                                                                     (args.max_speedup - 1) * 100, to_send),
+          'Sections kept in %s so a re-run only sends what changed. Remove that folder when the video is final.' % work])
+    return OK
+
+
+def cmd_polish(args):
+    check_inputs(args)
+    need_tools()
+    length, take = duration(args.video), duration(args.audio)
+    if not length or not take:
+        raise Stop('Cannot read the duration of the video or the recording.')
+    if args.isolate:
+        key()
+    work = work_dir(args)
+    plan = {'audio': args.audio, 'take_seconds': take, 'video': args.video, 'video_seconds': length, 'out': args.out, 'isolate': args.isolate,
+            'normalize': args.normalize, 'trim': take > length + 0.5, 'pad': take < length - 0.5}
+    if args.isolate and not args.confirm:
+        emit(args, dict(plan, confirmed=False), ['Plan only, nothing was sent.', 'The take (%.1fs) would be cleaned by ElevenLabs, then levelled and placed on the %.1fs video.' % (take, length),
+                                                  'Run again with --confirm.'])
+        return PLAN_ONLY
+    ensure_free([args.out, track_of(args.out)], args.force)
+    os.makedirs(work, exist_ok=True)
+    source = args.audio
+    if args.isolate:
+        cleaned = os.path.join(work, 'take-clean-%s.mp3' % hashlib.sha256(read_bytes(args.audio)).hexdigest()[:12])
+        if not os.path.isfile(cleaned):
+            body, content_type = multipart({}, [('audio', os.path.basename(args.audio), read_bytes(args.audio))])
+            status, _, data = call('POST', '/v1/audio-isolation', body=body, content_type=content_type)
+            if status != 200 or not data:
+                raise Stop('Isolation failed. ' + explain(status, data))
+            write_atomic(cleaned, data, True)
+        source = cleaned
+    filters = (['loudnorm=I=-16:TP=-1.5:LRA=11'] if args.normalize else []) + (['afade=t=out:st=%.3f:d=0.3' % (length - 0.3)] if take > length + 0.5 else [])
+    filters += ['apad=whole_dur=%.3f' % length, 'atrim=end=%.3f' % length]
+    mix = os.path.join(work, '_mix.wav')
+    run(['ffmpeg', '-y', '-loglevel', 'error', '-i', source, '-af', ','.join(filters), '-ar', '44100', '-ac', '2', mix])
+    track, poster = mux(args.video, mix, args.out, length, args.force)
+    emit(args, dict(plan, confirmed=True, out=args.out, track=track, poster=poster),
+         ['Wrote %s' % args.out, 'Voiceover track: %s' % track] + (['Trimmed the take to the video length.'] if plan['trim'] else []))
+    return OK
+
+
+# ---- command line
+
+def text_arg(value):
+    if re.search(r'["$`\\\n]', value):
+        raise argparse.ArgumentTypeError('may not contain quotes, $, backticks or backslashes')
+    return value
+
+
+def build_parser():
+    parser = argparse.ArgumentParser(prog='elevenlabs.py', description='ElevenLabs for the final voiceover of a narrated video.')
+    sub = parser.add_subparsers(dest='command', required=True)
+
+    def common(p, voice=False, confirm=False, out=False, work=False):
+        p.add_argument('--json', action='store_true', help='print JSON')
+        if voice:
+            p.add_argument('--voice', type=text_arg, default=os.environ.get('ELEVENLABS_VOICE', ''), help='voice id')
+            p.add_argument('--model', type=text_arg, default=os.environ.get('ELEVENLABS_MODEL') or DEFAULT_MODEL)
+            p.add_argument('--format', type=text_arg, default=DEFAULT_FORMAT)
+        if confirm:
+            p.add_argument('--confirm', action='store_true', help='actually send; without it only the plan is printed')
+        if out:
+            p.add_argument('--out', type=text_arg, required=True)
+            p.add_argument('--force', action='store_true', help='replace an existing file')
+        if work:
+            p.add_argument('--work', type=text_arg, default='', help='where the per-section audio is kept')
+
+    p = sub.add_parser('sections', help='the timed sections of a voiceover script')
+    p.add_argument('--script', type=text_arg, required=True)
+    p.add_argument('--drop-optional', action='store_true')
+    common(p)
+    p = sub.add_parser('voices', help='list the voices on the account')
+    p.add_argument('--search', type=text_arg, default='')
+    p.add_argument('--limit', type=int, default=30)
+    common(p)
+    p = sub.add_parser('account', help='plan and characters left')
+    common(p)
+    p = sub.add_parser('speak', help='one text file to one audio file')
+    p.add_argument('--text-file', type=text_arg, required=True)
+    common(p, voice=True, confirm=True, out=True)
+    p = sub.add_parser('isolate', help='remove background noise from a recording')
+    p.add_argument('--audio', type=text_arg, required=True)
+    common(p, confirm=True, out=True)
+    p = sub.add_parser('voiceover', help='speech for every section, fitted to its window, on the approved draft')
+    p.add_argument('--script', type=text_arg, required=True)
+    p.add_argument('--video', type=text_arg, required=True)
+    p.add_argument('--drop-optional', action='store_true')
+    p.add_argument('--max-speedup', type=float, default=1.08)
+    p.add_argument('--redo', action='store_true', help='ignore the sections already made')
+    common(p, voice=True, confirm=True, out=True, work=True)
+    p = sub.add_parser('polish', help='level a recorded take and put it on the approved draft')
+    p.add_argument('--audio', type=text_arg, required=True)
+    p.add_argument('--video', type=text_arg, required=True)
+    p.add_argument('--isolate', action='store_true', help='clean the take with ElevenLabs first')
+    p.add_argument('--no-normalize', dest='normalize', action='store_false')
+    common(p, confirm=True, out=True, work=True)
+    return parser
+
+
+COMMANDS = {'sections': cmd_sections, 'voices': cmd_voices, 'account': cmd_account, 'speak': cmd_speak, 'isolate': cmd_isolate,
+            'voiceover': cmd_voiceover, 'polish': cmd_polish}
+
+
+def main(argv):
+    args = build_parser().parse_args(argv)
+    if args.command in ('speak', 'voiceover'):
+        if not args.voice:
+            print('No voice. Pass --voice <id>, or set voice_id in settings.yml (list them: -a voices).', file=sys.stderr)
+            return USAGE
+        audio_ext(args.format)
+    try:
+        if args.command in ('speak', 'voiceover'):
+            args.voice = resolve_voice(args.voice)
+        return COMMANDS[args.command](args)
+    except Stop as stop:
+        print(stop, file=sys.stderr)
+        return stop.code
+    except KeyboardInterrupt:
+        print('Interrupted. Sections already made are kept.', file=sys.stderr)
+        return ERROR
+
+
+if __name__ == '__main__':
+    sys.exit(main(sys.argv[1:]))

package/core/scripts/libs/progress-bar.sh

@@ -0,0 +1,19 @@
+#!/usr/bin/env bash
+
+# 1. Create ProgressBar function
+# 1.1 Input is currentState($1) and totalState($2)
+function progress_bar {
+# Process data
+    let _progress=(${1}*100/${2}*100)/100
+    let _done=(${_progress}*4)/10
+    let _left=40-$_done
+# Build progressbar string lengths
+    _fill=$(printf "%${_done}s")
+    _empty=$(printf "%${_left}s")
+
+# 1.2 Build progressbar strings and print the ProgressBar line
+# 1.2.1 Output example:                           
+# 1.2.1.1 Progress : [########################################] 100%
+printf "\rProgress : [${_fill// /\#}${_empty// /-}] ${_progress}%%\n"
+
+}