bol

Speech Recognition Library for Indic Languages

Installation (Currently in Alpha)

sudo apt-get install sox libsox-fmt-mp3 libsndfile1

pip install --upgrade bol-library==0.1.8

Do Speech Recognition in 3 lines of code

Usage:

from bol.models import load_model

model = load_model('hi-ts')

text = model.predict('/home/harveen/bol/dev/long/virat.wav')   

Output:

मेरा नाम विराट कोहली है 

Model Zoo

Proudly hosted on HuggingFace Datasets

Unique Code	Language Code	Backend	Algo	Language	provider	LM Support	CPU	GPU
hi-ts	hi-IN	torchscript	wav2vec2	Hindi	vakyansh_ekstep	✖️	✔️	✖️
bn-ts	bn-IN	torchscript	wav2vec2	Bengali	vakyansh_ekstep	✖️	✔️	✖️
gn-ts	gn-IN	torchscript	wav2vec2	Gujarati	vakyansh_ekstep	✖️	✔️	✖️
en-in-ts	en-IN	torchscript	wav2vec2	Indian English	vakyansh_ekstep	✖️	✔️	✖️
kn-ts	kn-IN	torchscript	wav2vec2	Kannada	vakyansh_ekstep	✖️	✔️	✖️
ne-ts	ne-IN	torchscript	wav2vec2	Nepali	vakyansh_ekstep	✖️	✔️	✖️
ta-ts	ta-IN	torchscript	wav2vec2	Tamil	vakyansh_ekstep	✖️	✔️	✖️
te-ts	te-IN	torchscript	wav2vec2	Telugu	vakyansh_ekstep	✖️	✔️	✖️
hi-vakyansh	hi-IN	fairseq	wav2vec2	Hindi	vakyansh_ekstep	✔️	✔️	✔️
bn-vakyansh	bn-IN	fairseq	wav2vec2	Bengali	vakyansh_ekstep	✔️	✔️	✔️
en-vakyansh	en-IN	fairseq	wav2vec2	English	vakyansh_ekstep	✔️	✔️	✔️
gu-vakyansh	gu-IN	fairseq	wav2vec2	Gujarati	vakyansh_ekstep	✔️	✔️	✔️
kn-vakyansh	kn-IN	fairseq	wav2vec2	Kannada	vakyansh_ekstep	✔️	✔️	✔️
ne-vakyansh	ne-IN	fairseq	wav2vec2	Nepali	vakyansh_ekstep	✔️	✔️	✔️
ta-vakyansh	ta-IN	fairseq	wav2vec2	Tamil	vakyansh_ekstep	✔️	✔️	✔️
te-vakyansh	te-IN	fairseq	wav2vec2	Telugu	vakyansh_ekstep	✔️	✔️	✔️

Project Vision

1. Have state of the art speech recognition models for all the indic languages with or without language model

2. Provide High Performance Inference Pipelines for various backbones.

3. Provide Inference Pipeline for Speech Translation, Speech to Text Translation, Multilingual ASR