Add Per-Stream Phrase Boosting for Cache-Aware RNN-T

[artbataev]

Important

The Update branch button must only be pressed in very rare occassions.
An outdated branch is never blocking the merge of a PR.
Please reach out to the automation team before pressing that button.

What does this PR do ?

Support per-stream biasing in inference for cache-aware models.
Per-stream biasing was introduced in #15125

Experiments

(1) Global vs Per-Stream Boosting

• Model: private cache-aware without PnC
• Dataset: private (for phrase boosting)

Boosting	Cache	WERR (%)	RTFx
no (baseline)	-	baseline	598.08
global	-	6.2%	548.35
per-stream	no	6.2%	114.46
per-stream	memory	6.2%	492.91

WERR for global and per-stream boosting is the same.
Memory cache significantly improves the speed as expected.

(2) Sanity check with boosting ground truth on LibriSpeech

• Model: private cache-aware without PnC
• Dataset: LibriSpeech test-other

Boosting	Cache	WER (%)	RTFx
no	-	5.94	610.79
Per-stream (ground truth)	no	1.26	451.41
Per-stream (ground truth)	memory	1.26	565.61

Boosting ground truth significantly improves WER as expected.

Collection: [ASR]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

python examples/asr/asr_streaming_inference/asr_streaming_infer.py \
    --config-path="../conf/asr_streaming_inference/" \
    --config-name=cache_aware_rnnt.yaml \
    audio_file="<manifest_with_boosting_requests>" \
    output_filename="result.jsonl" \
    asr.model_name="nvidia/nemotron-speech-streaming-en-0.6b" \
    streaming.batch_size=5 \
    lang=en \
    enable_pnc=False \
    enable_itn=False \
    enable_nmt=False \
    asr_output_granularity=segment

Manifest example:

{"audio_filepath": "/data/file1.wav", "text": "some ground truth text", "biasing_request": {"boosting_model_cfg": {"key_phrases_list": ["one phrase"]}}}
{"audio_filepath": "/data/file2.wav", "text": "another ground truth text", "biasing_request": {"boosting_model_cfg": {"key_phrases_list": ["other phrases", "and this one"]}}}

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[github-actions]

[🤖]: Hi @artbataev 👋,

We wanted to let you know that a CICD pipeline for this PR just finished successfully.

So it might be time to merge this PR or get some approvals.

//cc @chtruong814 @ko3n1g @pablo-garay @thomasdhc

[naymaraq]

Everything looks great