Hotfix: add missing apply_whisper_normalization

[Jorjeous]

add missing apply_whisper_normalization attribute to AudioEvaluator

Summary by CodeRabbit

• New Features
  • Added a new configuration option for audio evaluation that provides control over Whisper-based normalization. Users can enable or disable this feature as needed; normalization is enabled by default. When disabled, alternate preprocessing modes are applied to evaluation data during scoring.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[Jorjeous]

Some merge conflict was resolved incorrectly, fixing this

[coderabbitai]

📝 Walkthrough

Walkthrough

Added a new boolean configuration flag apply_whisper_normalization (default True) to the AudioEvaluatorConfig dataclass. When disabled, this flag causes normalization_mode to be set to "none" during evaluation preprocessing, affecting how ASR-PC and related evaluations handle reference and hypothesis normalization.

Changes

Cohort / File(s)	Summary
Audio evaluator configuration nemo_skills/evaluation/evaluator/audio.py	Added public dataclass field apply_whisper_normalization: bool = True to AudioEvaluatorConfig. Controls whether Whisper-based normalization is applied during ASR evaluation; when False, normalization_mode defaults to "none".

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Title check	✅ Passed	The title accurately identifies the main change: adding the missing apply_whisper_normalization attribute to AudioEvaluatorConfig.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[greptile-apps]

Greptile Overview

Greptile Summary

This PR fixes a bug introduced in PR #1157 where the apply_whisper_normalization attribute was used in the evaluate_sample function but was never defined in the AudioEvaluatorConfig class.

• Added missing apply_whisper_normalization: bool = True to AudioEvaluatorConfig class definition
• The attribute is referenced on lines 512, 522, and 529 to control whether Whisper normalization should be applied during ASR evaluation
• Without this fix, attempting to evaluate audio samples would result in an AttributeError when the config tries to access this undefined attribute
• The default value of True maintains backward compatibility and expected behavior

Confidence Score: 5/5

• This PR is safe to merge with no risk
• The change is a simple one-line addition that fixes a clear bug where an attribute was being accessed but never defined. The fix is straightforward, uses the correct type and default value, and aligns perfectly with how the attribute is used throughout the codebase.
• No files require special attention

Important Files Changed

Filename	Overview
nemo_skills/evaluation/evaluator/audio.py	Added missing apply_whisper_normalization attribute to AudioEvaluatorConfig - fixes AttributeError when config is accessed in evaluate_sample

Sequence Diagram

sequenceDiagram
    participant Client
    participant AudioEvaluator
    participant AudioEvaluatorConfig
    participant evaluate_sample
    participant evaluate_asr_pc
    
    Client->>AudioEvaluator: eval_single(data_point)
    AudioEvaluator->>AudioEvaluatorConfig: Initialize config
    Note over AudioEvaluatorConfig: apply_whisper_normalization: bool = True
    AudioEvaluator->>evaluate_sample: evaluate_sample(sample, config)
    
    alt task_type == "ASR-PC"
        evaluate_sample->>AudioEvaluatorConfig: config.apply_whisper_normalization
        Note over evaluate_sample: mode = normalization_mode if apply_whisper_normalization else "none"
        evaluate_sample->>evaluate_asr_pc: evaluate_asr_pc(reference, hypothesis, mode)
        evaluate_asr_pc-->>evaluate_sample: metrics
    end
    
    alt task_type == "ASR"
        evaluate_sample->>AudioEvaluatorConfig: config.apply_whisper_normalization
        Note over evaluate_sample: mode = normalization_mode if apply_whisper_normalization else "none"
        evaluate_sample->>evaluate_asr_pc: evaluate_asr(reference, hypothesis, mode)
        evaluate_asr_pc-->>evaluate_sample: metrics
    end
    
    alt task_type == "ASR_LEADERBOARD"
        evaluate_sample->>AudioEvaluatorConfig: config.apply_whisper_normalization
        Note over evaluate_sample: mode = normalization_mode if apply_whisper_normalization else "none"
        evaluate_sample->>evaluate_asr_pc: evaluate_asr(reference, hypothesis, mode)
        evaluate_asr_pc-->>evaluate_sample: metrics
    end
    
    evaluate_sample-->>AudioEvaluator: updates
    AudioEvaluator-->>Client: evaluation results

Loading

[greptile-apps]

_{No files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}