Model Overview

This collection contains large size versions of Conformer-RNNT (around 120M parameters) trained on MCV-10-Be with around 500 hours of belarusian speech. The model transcribes speech in lower case Belarusian alphabet, note that 'i' is differs from english 'i'.

Trained or fine-tuned NeMo models (with the file extenstion .nemo) can be converted to Riva models (with the file extension .riva) and then deployed. Here is a pre-trained Conformer-RNNT speech-to-text (STT) -- a.k.a. automatic speech recognition (ASR) -- Riva model.

Model Architecture

Conformer-RNNT model is a non-autoregressive variant of Conformer model [1] for Automatic Speech Recognition which uses RNNT loss/decoding instead of Transducer. You may find more info on the detail of this model here: Conformer-RNNT Model.

Training

The NeMo toolkit [3] was used for training the models for over several dozen epochs. These model are trained with this example script and this base config.

The tokenizers for these models were built using the text transcripts of the train set with this script.

As start point pretrained "STT En conformer-RNNT Large" was taken, and then fine-tuned to Belarusian.

Datasets

Model was trined on MCV-10-Be dataset with 465h for train, 26 on test, 25 on dev.

Performance

Performances of the ASR models are reported in terms of Word Error Rate (WER%) with greedy decoding. WER on dev is 3.8%

How to Use this Model

The model is available for use in the NeMo toolkit [3], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Automatically load the model from NGC

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.EncDecCTCModelBPE.from_pretrained(model_name="stt_be_conformer_transducer_large")

Transcribing text with this model

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py \
 pretrained_name="stt_be_conformer_transducer_large" \
 audio_dir=""

Input

This model accepts 16000 Hz Mono-channel Audio (wav files) as input.

Output

This model provides transcribed speech as a string for a given audio sample.

Limitations

Since all models are trained on just MCV-10 dataset, the performance of this model might degrade for speech which includes technical terms, or vernacular that the model has not been trained on. The model might also perform worse for accented speech.

References

[1] Conformer: Convolution-augmented Transformer for Speech Recognition

[2] Google Sentencepiece Tokenizer

[3] NVIDIA NeMo Toolkit

Licence

License to use this model is covered by the NGC TERMS OF USE unless another License/Terms Of Use/EULA is clearly specified. By downloading the public and release version of the model, you accept the terms and conditions of the NGC TERMS OF USE.