refactor versa in OO (major update)

[ftshijt]

Refactor codebase for better interface:

• Introduce base metric for various setup
• Complete meta information within the metric setup
• MetricRegistry for automatic metric update
• MetricFactory for building metric
• MetricSuite for grouped modeling
• Improved dependency control

Fully supported metrics:

• asr_matching
• srmr
• asvspoof_score
• audiobox_aesthetics
• cdpam
• dpam
• chroma_alignment
• emo_similarity
• emo_vad
• nisqa
• discrete_speech
• owsm_lid
• noresqa

[Copilot]

Pull Request Overview

This pull request refactors the versa codebase to adopt an object‐oriented design for metrics, replacing function‐based implementations with class-based ones. Key changes include:

• Introducing class-based SRMRMetric and ASRMatchMetric along with their auto-registration functions.
• Updating the MetricRegistry, MetricFactory, and MetricSuite patterns to support the new design.
• Adapting the test pipelines and module initialization to load and use the new metric classes.

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
versa/utterance_metrics/srmr.py	Refactored SRMR metric into a class-based implementation with improved configuration.
versa/utterance_metrics/asr_matching.py	Replaced function-based ASR matching with a class-based version and updated its setup.
versa/metrics.py	Minor formatting adjustments to support new changes.
versa/definition.py	Added and updated abstract base classes, registry, factory, and suite for metrics.
versa/init.py	Updated imports to register and export the new metric classes instead of legacy functions.
test/test_pipeline/test_srmr.py	Adapted the SRMR test pipeline to use the registry and new scoring API.
test/test_pipeline/test_asr_match.py	Adapted the ASR match test pipeline to use the registry and new scoring API.

Comments suppressed due to low confidence (4)

versa/utterance_metrics/asr_matching.py:122

• The function 'opcodes' is used here but is not imported or defined in this module. Consider importing it from difflib (e.g., using SequenceMatcher.get_opcodes()) or defining it appropriately.

for op, ref_st, ref_et, inf_st, inf_et in opcodes(ref_chars, pred_chars):

versa/utterance_metrics/asr_matching.py:95

• The 'torch' module is used in this context but there is no corresponding import statement. Please add 'import torch' at the top of the module to ensure proper functionality.

with torch.no_grad():

versa/utterance_metrics/asr_matching.py:92

• The 'librosa' module is used for resampling operations in this file but is not imported. Please add 'import librosa' at the beginning of the file.

if fs != TARGET_FS:

versa/utterance_metrics/asr_matching.py:69

• The 'whisper' module is utilized to load the model, yet there is no import for it in the diff. Please add 'import whisper' at the top of the module to avoid runtime errors.

self.model = whisper.load_model(self.model_tag, device=self.device)