Skip to main content

Decodeur phonetique du francais — audio vers phones IPA structures (CNN-BiGRU-CTC)

Project description

lectura-ctc — Decodeur phonetique CTC du francais

Transcription audio vers phonemes IPA via un modele CNN-BiGRU-CTC medium (10.6M params, PER ~4.34%).

Le modele medium supporte les sigles, formules (nombres, dates, calculs) et la parole courante grace a un vocabulaire etendu de 59 tokens et un fine-tuning specialise.

Installation

# Avec backend ONNX (recommande)
pip install lectura-ctc[onnx]

# Sans ONNX (mode API uniquement)
pip install lectura-ctc

Utilisation

import numpy as np
from lectura_ctc import creer_engine

engine = creer_engine()

# Audio PCM float32 mono 16kHz
audio = np.zeros(16000, dtype=np.float32)  # 1 seconde de silence
result = engine.transcrire(audio)
print(result)  # chaine IPA : "b ɔ̃ ʒ u ʁ | l ə | m ɔ̃ d"

Modele

Caracteristique Valeur
Architecture CNN [48, 96] + BiGRU 384x4 + CTC head
Parametres 10.6M
Performance PER ~4.34% (formules v2)
Vocabulaire 59 tokens (46 phones IPA + liaisons + ponctuation + speciaux)
Entrainement Pre-train corpus complet + fine-tune formules/sigles multispeaker

Backends

Backend Dependance Latence Modele
ONNX Runtime onnxruntime ~10 ms/s audio phone_ctc_int8.onnx (38 Mo)
API aucune ~100 ms/s audio serveur Lectura

Parametres audio

  • Sample rate : 16 kHz
  • Format : PCM float32 mono
  • Mel : 80 bins, n_fft=512, hop=160, win=400

Licence

Double licence : AGPL-3.0 (libre) + Licence Commerciale (payante).

Les modeles ONNX sont distribues separement — voir MODEL_LICENCE.md.

Auteur

Max Carriere — lectura.world

Project details


Download files

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

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

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

lectura_decodeur-3.0.1-py3-none-any.whl (31.8 kB view details)

Uploaded Python 3

File details

Details for the file lectura_decodeur-3.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for lectura_decodeur-3.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 0b22503e232acd9175e4a0b52401bb84444380182fd85039dab917b9d4b739c6
MD5 3a95a97e9d26a8dbc93de8c2e4a0ccd2
BLAKE2b-256 ce04c67dab828c88c320e00e7b44aa06e56d643d6cdc188bd45183619ee8c7f7

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