Skip to main content

Synthese vocale neuronale multi-speaker francais — Matcha-Conformer + HiFi-GAN (ONNX)

Project description

lectura-multispeaker

Moteur de synthese vocale neuronale multi-speaker pour le francais — Matcha-Conformer + HiFi-GAN (ONNX).

Supporte 6 voix : siwis, ezwa, nadine, bernard, gilles, zeckou.

Installation

# Version minimale (API distante, zero deps)
pip install lectura-multispeaker

# Version locale (inference ONNX)
pip install lectura-multispeaker[onnx]

Pour le pipeline complet texte → audio, utilisez pip install lectura-tts-multi (inclut le G2P).

Utilisation

Depuis des phonemes IPA

from lectura_multispeaker import creer_engine

engine = creer_engine(mode="local")
engine.set_speaker("siwis")
result = engine.synthesize_phonemes(
    "bɔ̃ʒuʁ",
    phrase_type=0,
)
# result.samples: numpy float32
# result.sample_rate: 22050
# result.phoneme_timings: list[PhonemeTiming]

Lister les speakers disponibles

from lectura_multispeaker import liste_speakers
print(liste_speakers())  # ['siwis', 'ezwa', 'nadine', 'bernard', 'gilles', 'zeckou']

Controles prosodiques

Parametre Defaut Description
duration_scale 1.0 Vitesse globale
pitch_shift 0.0 Decalage F0 (demi-tons)
pitch_range 1.0 Variation F0
energy_scale 1.0 Intensite
pause_scale 1.0 Duree des pauses
phrase_type 0 0=decl, 1=inter, 2=excl, 3=susp
n_ode_steps 4 Pas ODE Matcha (plus = meilleur)

Architecture

  • Matcha-Conformer : phonemes → mel-spectrogramme via flow matching ODE (encodeur par speaker)
  • HiFi-GAN V1 : mel → audio 22050 Hz
  • Runtime : ONNX (pas de dependance PyTorch)

Emplacements des modeles

Recherche dans l'ordre :

  1. Parametre models_dir explicite
  2. $LECTURA_MODELS_DIR/tts_multispeaker/
  3. ~/.lectura/models/tts_multispeaker/
  4. Modeles embarques dans le package (version privee)

Licence

Licence proprietaire Lectura. Voir LICENCE.txt.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

lectura_multispeaker-3.1.0.tar.gz (28.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

lectura_multispeaker-3.1.0-py3-none-any.whl (25.7 kB view details)

Uploaded Python 3

File details

Details for the file lectura_multispeaker-3.1.0.tar.gz.

File metadata

  • Download URL: lectura_multispeaker-3.1.0.tar.gz
  • Upload date:
  • Size: 28.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.11.9

File hashes

Hashes for lectura_multispeaker-3.1.0.tar.gz
Algorithm Hash digest
SHA256 06375e191685adf76a80df883413937dc138a2f8be690ed1c33bb2d774a15177
MD5 85ffcb01ec0870bccf794066e36320c9
BLAKE2b-256 b4502aa9f58aafc4f27b2006ed32c970bf9c843f7230ba237f8ddb7809c59648

See more details on using hashes here.

File details

Details for the file lectura_multispeaker-3.1.0-py3-none-any.whl.

File metadata

File hashes

Hashes for lectura_multispeaker-3.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 0a6139f67c423b6ebd8db17ba72779ee05402099be7623fb83ea77409d099b2f
MD5 34b60d9d4320d4ca174d8a3687a60dab
BLAKE2b-256 0d12feafd586d5f453c7be151ffc9c1b28b066eef924bd097fd3cb5cd77e4c9d

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page