russell@unturf.com
02b4e7aaf7
Standard voices: alloy, echo, fable, onyx, nova, shimmer
Extended voices: amber, breeze, coral, dawn, ember, frost,
glow, haze, ivy, jade, kite, lark, mist, nectar
Source: LJ Speech Dataset (public domain)
215 lines
7.7 KiB
Python
215 lines
7.7 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
Download diverse voice samples for Qwen3-TTS voice cloning.
|
|
Uses HuggingFace datasets for LibriSpeech samples with multiple speakers.
|
|
"""
|
|
|
|
import os
|
|
import json
|
|
import argparse
|
|
from pathlib import Path
|
|
|
|
try:
|
|
from datasets import load_dataset
|
|
import soundfile as sf
|
|
HAS_DATASETS = True
|
|
except ImportError:
|
|
HAS_DATASETS = False
|
|
print("Install datasets library: pip install datasets soundfile")
|
|
|
|
# LibriSpeech test-clean speaker metadata
|
|
# Format: speaker_id: (gender, description)
|
|
LIBRISPEECH_SPEAKERS = {
|
|
# Female speakers
|
|
"1089": ("female", "clear, professional"),
|
|
"1188": ("female", "warm, measured"),
|
|
"1221": ("female", "expressive, storyteller"),
|
|
"1320": ("female", "soft, gentle"),
|
|
"3570": ("female", "bright, energetic"),
|
|
"3575": ("female", "calm, neutral"),
|
|
"4446": ("female", "mature, authoritative"),
|
|
"4507": ("female", "young, clear"),
|
|
"5142": ("female", "warm, friendly"),
|
|
"6829": ("female", "crisp, articulate"),
|
|
"6930": ("female", "smooth, radio-like"),
|
|
"7729": ("female", "light, airy"),
|
|
"8230": ("female", "rich, full"),
|
|
"8455": ("female", "neutral, newsreader"),
|
|
|
|
# Male speakers
|
|
"1284": ("male", "deep, resonant"),
|
|
"1580": ("male", "clear, narrator"),
|
|
"2094": ("male", "warm, baritone"),
|
|
"2830": ("male", "young, energetic"),
|
|
"4077": ("male", "mature, professor"),
|
|
"4970": ("male", "smooth, announcer"),
|
|
"5105": ("male", "neutral, clear"),
|
|
"5639": ("male", "deep, dramatic"),
|
|
"7021": ("male", "light, conversational"),
|
|
"7127": ("male", "authoritative, news"),
|
|
"7176": ("male", "warm, storyteller"),
|
|
"8224": ("male", "crisp, professional"),
|
|
"8463": ("male", "rich, bass"),
|
|
}
|
|
|
|
# Voice name mappings for OpenAI-compatible names
|
|
VOICE_MAPPINGS = {
|
|
# OpenAI standard voices
|
|
"alloy": {"speaker": "1089", "desc": "neutral, balanced female"},
|
|
"echo": {"speaker": "1284", "desc": "deep, clear male"},
|
|
"fable": {"speaker": "1221", "desc": "expressive, storyteller female"},
|
|
"onyx": {"speaker": "5639", "desc": "deep, dramatic male"},
|
|
"nova": {"speaker": "1188", "desc": "warm, friendly female"},
|
|
"shimmer": {"speaker": "1320", "desc": "soft, gentle female"},
|
|
|
|
# Extended voices - female
|
|
"aurora": {"speaker": "3570", "desc": "bright, energetic female"},
|
|
"bella": {"speaker": "4446", "desc": "mature, authoritative female"},
|
|
"clara": {"speaker": "5142", "desc": "warm, conversational female"},
|
|
"dawn": {"speaker": "6829", "desc": "crisp, articulate female"},
|
|
"ember": {"speaker": "6930", "desc": "smooth, radio-like female"},
|
|
"fiona": {"speaker": "7729", "desc": "light, airy female"},
|
|
"grace": {"speaker": "8230", "desc": "rich, full female"},
|
|
"hazel": {"speaker": "8455", "desc": "neutral, newsreader female"},
|
|
"iris": {"speaker": "3575", "desc": "calm, neutral female"},
|
|
"jade": {"speaker": "4507", "desc": "young, clear female"},
|
|
|
|
# Extended voices - male
|
|
"atlas": {"speaker": "1580", "desc": "clear, narrator male"},
|
|
"blaze": {"speaker": "2830", "desc": "young, energetic male"},
|
|
"cedar": {"speaker": "4077", "desc": "mature, professor male"},
|
|
"drake": {"speaker": "4970", "desc": "smooth, announcer male"},
|
|
"eric": {"speaker": "5105", "desc": "neutral, clear male"},
|
|
"felix": {"speaker": "7021", "desc": "light, conversational male"},
|
|
"grant": {"speaker": "7127", "desc": "authoritative, news male"},
|
|
"hugo": {"speaker": "7176", "desc": "warm, storyteller male"},
|
|
"ivan": {"speaker": "8224", "desc": "crisp, professional male"},
|
|
"jack": {"speaker": "2094", "desc": "warm, baritone male"},
|
|
"knox": {"speaker": "8463", "desc": "rich, bass male"},
|
|
}
|
|
|
|
|
|
def download_librispeech_samples(output_dir: Path, max_per_speaker: int = 1):
|
|
"""Download LibriSpeech test-clean samples for each speaker."""
|
|
|
|
if not HAS_DATASETS:
|
|
print("Error: datasets library not installed")
|
|
return {}
|
|
|
|
print("Loading LibriSpeech test-clean dataset...")
|
|
dataset = load_dataset(
|
|
"openslr/librispeech_asr",
|
|
"clean",
|
|
split="test",
|
|
trust_remote_code=True
|
|
)
|
|
|
|
# Group samples by speaker
|
|
speaker_samples = {}
|
|
for sample in dataset:
|
|
speaker_id = str(sample["speaker_id"])
|
|
if speaker_id not in speaker_samples:
|
|
speaker_samples[speaker_id] = []
|
|
if len(speaker_samples[speaker_id]) < max_per_speaker:
|
|
speaker_samples[speaker_id].append(sample)
|
|
|
|
# Download samples for our target speakers
|
|
downloaded = {}
|
|
output_dir.mkdir(parents=True, exist_ok=True)
|
|
|
|
for voice_name, voice_info in VOICE_MAPPINGS.items():
|
|
speaker_id = voice_info["speaker"]
|
|
|
|
if speaker_id not in speaker_samples:
|
|
print(f" Warning: Speaker {speaker_id} not found for voice '{voice_name}'")
|
|
continue
|
|
|
|
sample = speaker_samples[speaker_id][0]
|
|
audio = sample["audio"]
|
|
transcript = sample["text"]
|
|
|
|
# Save audio file
|
|
output_path = output_dir / f"{voice_name}.wav"
|
|
sf.write(str(output_path), audio["array"], audio["sampling_rate"])
|
|
|
|
downloaded[voice_name] = {
|
|
"file": str(output_path),
|
|
"transcript": transcript,
|
|
"speaker_id": speaker_id,
|
|
"description": voice_info["desc"],
|
|
"sample_rate": audio["sampling_rate"]
|
|
}
|
|
|
|
print(f" Downloaded: {voice_name} (speaker {speaker_id})")
|
|
|
|
return downloaded
|
|
|
|
|
|
def generate_voice_config(voices: dict, output_file: Path):
|
|
"""Generate voice_to_speaker.yaml config."""
|
|
|
|
lines = [
|
|
"# uncloseai-speech Voice Configuration",
|
|
"# Auto-generated with diverse LibriSpeech speakers",
|
|
"",
|
|
"tts-1-qwen:",
|
|
" # OpenAI-compatible voice names with diverse speakers",
|
|
" # Each voice has a unique speaker from LibriSpeech test-clean",
|
|
"",
|
|
]
|
|
|
|
for voice_name, info in sorted(voices.items()):
|
|
# Use relative path from app root
|
|
rel_path = info["file"].replace("/app/", "").replace(str(Path.cwd()) + "/", "")
|
|
if not rel_path.startswith("voices/"):
|
|
rel_path = f"voices/samples/{voice_name}.wav"
|
|
|
|
lines.extend([
|
|
f" # {info['description']}",
|
|
f" {voice_name}:",
|
|
f" ref_audio: {rel_path}",
|
|
f' ref_text: "{info["transcript"]}"',
|
|
f" language: English",
|
|
"",
|
|
])
|
|
|
|
with open(output_file, "w") as f:
|
|
f.write("\n".join(lines))
|
|
|
|
print(f"\nGenerated config: {output_file}")
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser(description="Download diverse voice samples")
|
|
parser.add_argument("--output-dir", "-o", default="voices/samples",
|
|
help="Output directory for voice samples")
|
|
parser.add_argument("--config-output", "-c", default="config/voice_to_speaker.yaml",
|
|
help="Output path for voice config")
|
|
parser.add_argument("--max-per-speaker", "-m", type=int, default=1,
|
|
help="Max samples per speaker")
|
|
args = parser.parse_args()
|
|
|
|
output_dir = Path(args.output_dir)
|
|
|
|
print(f"Downloading voice samples to: {output_dir}")
|
|
voices = download_librispeech_samples(output_dir, args.max_per_speaker)
|
|
|
|
if voices:
|
|
print(f"\nDownloaded {len(voices)} voice samples")
|
|
|
|
# Generate config
|
|
config_path = Path(args.config_output)
|
|
config_path.parent.mkdir(parents=True, exist_ok=True)
|
|
generate_voice_config(voices, config_path)
|
|
|
|
# Also save metadata
|
|
metadata_path = output_dir / "voices_metadata.json"
|
|
with open(metadata_path, "w") as f:
|
|
json.dump(voices, f, indent=2)
|
|
print(f"Saved metadata: {metadata_path}")
|
|
else:
|
|
print("No voices downloaded")
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|