Migrate obsoleted editdistance to rapidfuzz by excosy · Pull Request #3105 · modelscope/FunASR

excosy · 2026-07-02T14:49:35Z

Summary

Since editdistance has been archived and cannot be newly installed on Windows, replace it with rapidfuzz.

Type of change

Validation

python -m compileall funasr examples tests
Docs or links checked
Runtime/deployment command tested

gemini-code-assist

Code Review

This pull request replaces the editdistance dependency with rapidfuzz to calculate Character Error Rate (CER) and Word Error Rate (WER) in funasr/metrics/common.py, updating setup.py accordingly. The review feedback suggests addressing potential ZeroDivisionError bugs in calculate_cer and calculate_wer when reference sequences are empty. Additionally, it recommends reordering imports in common.py to adhere to PEP 8 standards and specifying a minimum version constraint of rapidfuzz>=3.0.0 in setup.py to prevent import errors.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

gemini-code-assist · 2026-07-02T14:51:36Z

        char_eds, char_ref_lens = [], []
        for i, seq_hat_text in enumerate(seqs_hat):
            seq_true_text = seqs_true[i]
            hyp_chars = seq_hat_text.replace(" ", "")
            ref_chars = seq_true_text.replace(" ", "")
-            char_eds.append(editdistance.eval(hyp_chars, ref_chars))
+            char_eds.append(Levenshtein.distance(hyp_chars, ref_chars))
            char_ref_lens.append(len(ref_chars))
        return float(sum(char_eds)) / sum(char_ref_lens)


If seqs_true contains only empty strings or strings with only spaces, char_ref_lens will sum to 0, leading to a ZeroDivisionError. Adding a guard check to ensure the reference length is greater than 0 before division prevents potential runtime crashes.

Suggested change

char_eds, char_ref_lens = [], []

for i, seq_hat_text in enumerate(seqs_hat):

seq_true_text = seqs_true[i]

hyp_chars = seq_hat_text.replace(" ", "")

ref_chars = seq_true_text.replace(" ", "")

char_eds.append(editdistance.eval(hyp_chars, ref_chars))

char_eds.append(Levenshtein.distance(hyp_chars, ref_chars))

char_ref_lens.append(len(ref_chars))

return float(sum(char_eds)) / sum(char_ref_lens)

char_eds, char_ref_lens = [], []

for i, seq_hat_text in enumerate(seqs_hat):

seq_true_text = seqs_true[i]

hyp_chars = seq_hat_text.replace(" ", "")

ref_chars = seq_true_text.replace(" ", "")

char_eds.append(Levenshtein.distance(hyp_chars, ref_chars))

char_ref_lens.append(len(ref_chars))

ref_len = sum(char_ref_lens)

return float(sum(char_eds)) / ref_len if ref_len > 0 else 0.0

gemini-code-assist · 2026-07-02T14:51:36Z

        word_eds, word_ref_lens = [], []
        for i, seq_hat_text in enumerate(seqs_hat):
            seq_true_text = seqs_true[i]
            hyp_words = seq_hat_text.split()
            ref_words = seq_true_text.split()
-            word_eds.append(editdistance.eval(hyp_words, ref_words))
+            word_eds.append(Levenshtein.distance(hyp_words, ref_words))
            word_ref_lens.append(len(ref_words))
        return float(sum(word_eds)) / sum(word_ref_lens)


If seqs_true contains only empty strings or strings with only spaces, word_ref_lens will sum to 0, leading to a ZeroDivisionError. Adding a guard check to ensure the reference length is greater than 0 before division prevents potential runtime crashes.

Suggested change

word_eds, word_ref_lens = [], []

for i, seq_hat_text in enumerate(seqs_hat):

seq_true_text = seqs_true[i]

hyp_words = seq_hat_text.split()

ref_words = seq_true_text.split()

word_eds.append(editdistance.eval(hyp_words, ref_words))

word_eds.append(Levenshtein.distance(hyp_words, ref_words))

word_ref_lens.append(len(ref_words))

return float(sum(word_eds)) / sum(word_ref_lens)

word_eds, word_ref_lens = [], []

for i, seq_hat_text in enumerate(seqs_hat):

seq_true_text = seqs_true[i]

hyp_words = seq_hat_text.split()

ref_words = seq_true_text.split()

word_eds.append(Levenshtein.distance(hyp_words, ref_words))

word_ref_lens.append(len(ref_words))

ref_len = sum(word_ref_lens)

return float(sum(word_eds)) / ref_len if ref_len > 0 else 0.0

gemini-code-assist · 2026-07-02T14:51:36Z

+from rapidfuzz.distance import Levenshtein
 from itertools import groupby


According to PEP 8, standard library imports (like itertools) should be grouped together and placed before third-party imports (like rapidfuzz). Please reorder these imports to adhere to PEP 8 guidelines.

Suggested change

from rapidfuzz.distance import Levenshtein

from itertools import groupby

from itertools import groupby

from rapidfuzz.distance import Levenshtein

References

PEP 8 states that imports should be grouped in the following order: 1. Standard library imports, 2. Related third party imports, 3. Local application/library specific imports. ^(link)

gemini-code-assist · 2026-07-02T14:51:36Z

        # Speaker & evaluation
        "umap_learn",
-        "editdistance>=0.5.2",
+        "rapidfuzz",


The rapidfuzz.distance module (specifically Levenshtein) was introduced in rapidfuzz version 2.0.0. To prevent potential ImportError issues on environments with older versions of rapidfuzz installed, please specify a minimum version constraint such as rapidfuzz>=3.0.0.

Suggested change

"rapidfuzz",

"rapidfuzz>=3.0.0",

gemini-code-assist · 2026-07-02T14:51:37Z

    # train: The modules invoked when training only.
    "train": [
-        "editdistance",
+        "rapidfuzz",


To remain consistent with the main requirements, please specify the same minimum version constraint for rapidfuzz in the train extra requirements.

Suggested change

"rapidfuzz",

"rapidfuzz>=3.0.0",

LauraGPT · 2026-07-07T06:38:19Z

Thanks for migrating away from archived editdistance; the direction makes sense, especially for Windows installs.

I validated current head 71af8eba locally and found one remaining issue from the high-priority review feedback: calculate_cer() / calculate_wer() still divide by zero when there is at least one hypothesis but every reference is empty or only spaces. The current if char_eds else None guard only handles empty input lists, not zero total reference length.

I prepared a minimal TDD patch on LauraGPT/FunASR:codex/fix-pr-3105-empty-ref-metrics:

adds regression coverage for empty-reference CER/WER
keeps normal Levenshtein semantics for non-empty references
returns None when total reference length is zero, matching calculate_cer_ctc()'s no-valid-reference behavior

Validation on ind-gpu8:

python -m pytest tests/test_metrics_common_rapidfuzz.py -q -> 2 passed
python -m py_compile funasr/metrics/common.py tests/test_metrics_common_rapidfuzz.py setup.py -> OK
git diff --check -> OK
temporary real rapidfuzz>=3.0.0 API probe confirms string and token-list distance semantics

Compare: https://github.com/modelscope/FunASR/compare/main...LauraGPT:FunASR:codex/fix-pr-3105-empty-ref-metrics?expand=1

LauraGPT

Thanks for taking on the editdistance -> rapidfuzz migration. The dependency direction is good, but current head 71af8eba still has a correctness blocker before merge.

calculate_cer() and calculate_wer() still divide by zero when the input list is non-empty but every reference is empty or whitespace-only. The current guard only checks whether there are edit-distance entries, not whether total reference length is zero.

I reproduced it with a focused regression test and prepared a small fix branch:

branch: LauraGPT/FunASR:codex/fix-pr-3105-empty-ref-metrics
compare: https://github.com/modelscope/FunASR/compare/main...LauraGPT:FunASR:codex/fix-pr-3105-empty-ref-metrics?expand=1

Validation on ind-gpu8:

red before fix: python -m pytest tests/test_metrics_common_rapidfuzz.py -q -> ZeroDivisionError
green after fix: python -m pytest tests/test_metrics_common_rapidfuzz.py -q -> 2 passed
python -m py_compile funasr/metrics/common.py tests/test_metrics_common_rapidfuzz.py setup.py -> OK
git diff --check -> OK
temporary real rapidfuzz>=3.0.0 API probe confirmed string and token-list distance semantics

Please adopt the zero-reference guard before this PR is merged. Returning None when total reference length is zero matches the existing CTC metric behavior.

Replace archived editdistance with rapidfuzz and return None when CER/WER references have zero total length. Supersedes #3105.

LauraGPT · 2026-07-07T07:00:54Z

Thanks again for starting the editdistance -> rapidfuzz migration.

I merged the maintainer-owned replacement in #3114 so we could include the empty-reference CER/WER guard before changing the dependency on main.

Your PR was the basis for the migration; the replacement keeps the same dependency direction and adds the missing zero-reference behavior:

calculate_cer(["abc"], [" "]) returns None
calculate_wer(["hello world"], [" "]) returns None

I am closing this PR to avoid having two open PRs for the same migration.

Migrate obsoleted editdistance to rapidfuzz

dad70eb

gemini-code-assist Bot reviewed Jul 2, 2026

View reviewed changes

Adopt gemini suggestions

71af8eb

LauraGPT requested changes Jul 7, 2026

View reviewed changes

LauraGPT mentioned this pull request Jul 7, 2026

Migrate editdistance metrics to rapidfuzz with empty-reference guard #3114

Merged

LauraGPT added a commit that referenced this pull request Jul 7, 2026

Migrate editdistance metrics to rapidfuzz with empty-reference guard

cd15a47

Replace archived editdistance with rapidfuzz and return None when CER/WER references have zero total length. Supersedes #3105.

LauraGPT closed this Jul 7, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Migrate obsoleted editdistance to rapidfuzz#3105

Migrate obsoleted editdistance to rapidfuzz#3105
excosy wants to merge 2 commits into
modelscope:mainfrom
excosy:main

excosy commented Jul 2, 2026 •

edited

Loading

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Uh oh!

LauraGPT commented Jul 7, 2026

Uh oh!

LauraGPT left a comment

Uh oh!

LauraGPT commented Jul 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		from rapidfuzz.distance import Levenshtein
		from itertools import groupby

Uh oh!

Conversation

excosy commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Type of change

Validation

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

LauraGPT commented Jul 7, 2026

Uh oh!

LauraGPT left a comment

Choose a reason for hiding this comment

Uh oh!

LauraGPT commented Jul 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

excosy commented Jul 2, 2026 •

edited

Loading