NVIDIA Streaming FastConformer-Hybrid Large (ka)

This collection contains large size versions of cache-aware FastConformer-Hybrid (around 115M parameters) trained on a Georgian speech. These models are trained for streaming ASR with look-ahead of 80ms which be used for very low-latency streaming applications. The model is hybrid with both Transducer and CTC decoders.

Model Architecture

These models are cache-aware versions of Hybrid FastConfomer which are trained for streaming ASR. You may find more info on cache-aware models here: Cache-aware Streaming Conformer [5]. The models are trained with multiple look-aheads which makes the model to be able to support different latencies. To learn on how to switch between different look-ahead, you may read the documentation on the cache-aware models.

FastConformer [4] is an optimized version of the Conformer model [1], and you may find more information on the details of FastConformer here: Fast-Conformer Model.

The model is trained in a multitask setup with joint Transducer and CTC decoder loss [5]. You can find more about Hybrid Transducer-CTC training here: Hybrid Transducer-CTC. You may also find more on how to switch between the Transducer and CTC decoders in the documentation.

NVIDIA NeMo: Training

The NeMo toolkit [3] was used for training the models for over several hundred epochs. These model are trained with this example script and this base config. The SentencePiece tokenizers [2] for these models were built using the text transcripts of the train set with this script.

Datasets

The model in this collection is trained on two datasets comprising approxinately 163 hours of Georgian speech:

• Mozilla Common Voice (v17.0) - 96 hours of validated data, 63 hours of unvalidated data
• Google Fleurs - 4 hours

Performance

The performance of Automatic Speech Recognition models is measuring using Word Error Rate. The following tables summarizes the performance of the available models in this collection with the Transducer decoder. Performances of the ASR models are reported in terms of Word Error Rate (WER%) with greedy decoding.

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 streaming or for fine-tuning on another dataset. You will need to install NVIDIA NeMo. We recommend you install it after you've installed latest Pytorch version.

pip install nemo_toolkit['all']

Automatically instantiate the model

import nemo.collections.asr as nemo_asr
asr_model = nemo_asr.models.EncDecHybridRNNTCTCBPEModel.from_pretrained(model_name="nvidia/stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc")

Transcribing using Python

Simply do:

output = asr_model.transcribe(['sample.wav'])
print(output[0].text)

Transcribing many audio files

Using Transducer mode inference:

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py
 pretrained_name="nvidia/stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc"
 audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"

Using CTC mode inference:

python [NEMO_GIT_FOLDER]/examples/asr/transcribe_speech.py
 pretrained_name="nvidia/stt_ka_fastconformer_hybrid_transducer_ctc_large_streaming_80ms_pc"
 audio_dir="<DIRECTORY CONTAINING AUDIO FILES>"
 decoder_type="ctc"

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 this model was trained on publically available speech datasets, 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.

NVIDIA Riva: Deployment

NVIDIA Riva, is an accelerated speech AI SDK deployable on-prem, in all clouds, multi-cloud, hybrid, on edge, and embedded. Additionally, Riva provides:

• World-class out-of-the-box accuracy for the most common languages with model checkpoints trained on proprietary data with hundreds of thousands of GPU-compute hours
• Best in class accuracy with run-time word boosting (e.g., brand and product names) and customization of acoustic model, language model, and inverse text normalization
• Streaming speech recognition, Kubernetes compatible scaling, and enterprise-grade support Although this model isn’t supported yet by Riva, the list of supported models is here. Check out Riva live demo.

References

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

[2] Google Sentencepiece Tokenizer

[3] NVIDIA NeMo Toolkit

[4] Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

[5] Stateful Conformer with Cache-based Inference for Streaming Automatic Speech Recognition

Licence

License to use this model is covered by the CC-BY-4.0. By downloading the public and release version of the model, you accept the terms and conditions of the CC-BY-4.0 license.