ASR inference: expose RNN-T decoding params for context biasing

[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An outdated branch is never blocking the merge of a PR.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 ?

Expose transducer decoding parameters (asr.decoding in transducer-related configs) to allow using context biasing in inference (boosting_tree and/or ngram_lm_model).

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=buffered_rnnt.yaml \
  asr.decoding.greedy.boosting_tree.key_phrases_file=${KEY_WORDS_LIST} \
  asr.decoding.greedy.boosting_tree_alpha=1.0 \
  audio_file=${DATASET} \
  streaming.batch_size=${BATCH_SIZE} \
  output_filename="${RESULTS_FILE}"

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The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.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

[Copilot]

Pull Request Overview

This PR exposes RNN-T transducer decoding parameters in streaming inference configs to enable context biasing features (boosting tree and n-gram language models) for ASR inference.

• Refactored RNNT decoding config initialization to merge user-provided configs with defaults
• Updated both buffered_pipeline_builder.py and cache_aware_pipeline_builder.py to accept and merge custom decoding parameters
• Added comprehensive decoding configuration sections to YAML config files exposing context biasing parameters

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
nemo/collections/asr/inference/factory/cache_aware_pipeline_builder.py	Refactored get_rnnt_decoding_cfg to accept config parameter and merge user settings with structured defaults
nemo/collections/asr/inference/factory/buffered_pipeline_builder.py	Refactored get_rnnt_decoding_cfg to accept config parameter and merge user settings with structured defaults
examples/asr/conf/asr_streaming_inference/cache_aware_rnnt.yaml	Added comprehensive decoding section with greedy strategy parameters, n-gram LM, and boosting tree configuration
examples/asr/conf/asr_streaming_inference/buffered_rnnt.yaml	Added comprehensive decoding section with greedy strategy parameters, n-gram LM, and boosting tree configuration, replacing previous top-level n-gram parameters

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[naymaraq]

Everything looks great.

[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