Skip to content

02-creating-profiles.md

Latest commit

 

History

History
178 lines (139 loc) · 4.97 KB

02-creating-profiles.md

File metadata and controls

178 lines (139 loc) · 4.97 KB
Raw

Creating Voice Profiles

Learn how to create voice profiles from audio samples.

Overview

A voice profile contains the audio samples and metadata needed to clone a voice. TTS Studio uses these profiles to generate speech that sounds like the original speaker.

Basic Profile Creation

from api.studio import TTSStudio

studio = TTSStudio()

# Create profile from audio samples
profile = studio.create_voice_profile(
    name="my_voice",
    sample_paths=[
        "./data/samples/neutral_01.wav",
        "./data/samples/happy_01.wav"
    ],
    language="es"
)

if profile["status"] == "success":
    print(f"Profile created: {profile['profile']['id']}")
else:
    print(f"Error: {profile['error']}")

Profile Creation Options

# Create profile with all options
profile = studio.create_voice_profile(
    name="my_voice",
    sample_paths=[
        "./data/samples/neutral_01.wav",
        "./data/samples/happy_01.wav",
        "./data/samples/serious_01.wav"
    ],
    language="es",  # Language code (es, en, fr, de, etc.)
    reference_text="Hola, esta es una muestra de mi voz para clonación."
)

Parameters:

  • name: Name for the voice profile (required)
  • sample_paths: List of paths to audio sample files (required)
  • language: Language code (default: "es")
  • reference_text: Optional text describing the samples

Audio Sample Requirements

For best results:

  • Duration: 3-30 seconds per sample
  • Quantity: 1-3 samples (Qwen3-TTS requires fewer samples)
  • Quality: Clear audio, minimal background noise
  • Format: WAV, 12000 Hz, mono, 16-bit (or will be converted)
  • Content: Natural speech, varied intonation
  • Consistency: Same speaker, similar recording conditions

Validating Samples

Before creating a profile, validate your samples:

# Validate samples first (optional but recommended)
validation = studio.validate_samples([
    "./data/samples/neutral_01.wav",
    "./data/samples/happy_01.wav"
])

if validation["all_valid"]:
    print(f"✓ All samples valid")
    print(f"Total duration: {validation['total_duration']}s")
else:
    print("✗ Some samples are invalid:")
    for result in validation["results"]:
        if not result["valid"]:
            print(f"  - {result['path']}: {result['error']}")

Complete Workflow

from api.studio import TTSStudio
from pathlib import Path

studio = TTSStudio()

# 1. Prepare audio samples
samples_dir = Path("./data/samples/my_voice")
sample_files = list(samples_dir.glob("*.wav"))

# 2. Validate samples first
validation = studio.validate_samples([str(f) for f in sample_files])

if validation["all_valid"]:
    # 3. Create the profile
    profile = studio.create_voice_profile(
        name="my_voice",
        sample_paths=[str(f) for f in sample_files],
        language="es"
    )

    if profile["status"] == "success":
        print(f"✓ Profile created: {profile['profile']['id']}")
        print(f"  Samples: {profile['profile']['num_samples']}")
        print(f"  Duration: {profile['profile']['total_duration']}s")
    else:
        print(f"✗ Error: {profile['error']}")
else:
    print("✗ Fix invalid samples before creating profile")

Managing Profiles

List All Profiles

# List all available profiles
profiles = studio.list_voice_profiles()

if profiles["status"] == "success":
    print(f"Found {profiles['count']} profiles:")
    for profile in profiles["profiles"]:
        print(f"  - {profile['name']} ({profile['id']})")
        print(f"    Samples: {profile['num_samples']}, Duration: {profile['total_duration']}s")

Delete a Profile

# Delete a profile by ID
result = studio.delete_voice_profile(profile_id="my_voice_20240115_143022")

if result["status"] == "success":
    print("✓ Profile deleted successfully")
else:
    print(f"✗ Error: {result['error']}")

Best Practices

For Profile Creation

  • Use high-quality audio recordings (12000 Hz, mono, 16-bit WAV)
  • Ensure consistent recording environment
  • Include varied speech patterns and emotions (1-3 samples)
  • Remove long silences and background noise
  • Use 3-30 seconds per sample

Sample Variety

  • Include different emotions (neutral, happy, serious)
  • Vary sentence structures (statements, questions)
  • Include different speaking speeds
  • Maintain consistent voice quality

Troubleshooting

Poor Voice Quality

  • Increase number of training samples (use 2-3 instead of 1)
  • Improve audio sample quality (remove noise, ensure clarity)
  • Check for background noise in samples
  • Ensure samples are 12000 Hz, mono, 16-bit

Profile Creation Fails

  • Validate samples first using validate_samples()
  • Check sample format (WAV, 12000 Hz, mono)
  • Ensure samples are 3-30 seconds each
  • Verify file paths are correct

Next Steps