WER
The primary evaluation metric in speech recognition, calculated as the sum of substitution, deletion, and insertion errors divided by the total number of words in the reference transcript. A WER below 5% is generally considered near human-level performance.
Core Facts
Word Error Rate (WER) is the most universal performance metric in the speech recognition field, calculated as (number of substitutions + number of deletions + number of insertions) / total number of words in the reference text.
Gemini 3.5 Transcribe is an extension of the Gemini series in the direction of speech processing, with two core capabilities of "precision" and "intelligence" as its main features.
Word Error Rate (WER) is calculated by dividing the sum of substitution, deletion, and insertion errors between the recognition output and the reference text by the total number of words in the reference text.
Timeline (last 90 days)
The formula for relative WER improvement is: (WER_SFT - WER_GRPO) / WER_SFT × 100%.
OCR commonly uses character error rate (CER) and word error rate (WER) as evaluation metrics; document AI focuses more on field-level accuracy and the TEDS score for table reconstruction.
The industry commonly uses speaker similarity (such as SECS) and word error rate (WER) as objective evaluation metrics for zero-shot speech cloning.
CER is more commonly used for languages without spaces such as Chinese, while WER is better suited to English.
Character Error Rate (CER) is defined as the edit distance (insertions + deletions + substitutions) divided by the total number of characters in the reference text; Word Error Rate (WER) is analogous but operates on words.
The WER formula is WER = (S + D + I) / N, where S is the number of substitutions, D the number of deletions, I the number of insertions, and N the total number of words in the reference transcription.
Evaluating speech translation quality usually uses metrics such as BLEU and COMET, which belong to a different evaluation system from ASR's WER.
Raw WER varies enormously across different datasets and different text normalization rules, so direct cross-vendor comparisons are prone to distortion.
In the end-to-end evaluation, the targeted repetition strategy brought a relative WER reduction of up to 24%.
WER as a release gate has two structural flaws: it is insensitive to semantics, and as a global average it cannot reflect how errors are distributed over key entities.
31 more timeline events
All Facts (20)
Word Error Rate (WER) is the most universal performance metric in the speech recognition field, calculated as (number of substitutions + number of deletions + number of insertions) / total number of words in the reference text.
90%VerifiedGemini 3.5 Transcribe is an extension of the Gemini series in the direction of speech processing, with two core capabilities of "precision" and "intelligence" as its main features.
85%VerifiedWord Error Rate (WER) is calculated by dividing the sum of substitution, deletion, and insertion errors between the recognition output and the reference text by the total number of words in the reference text.
80%VerifiedWord error rate (WER) is the core metric for measuring speech transcription accuracy, calculated as the sum of substitutions, deletions and insertions divided by the total number of words in the reference text.
75%VerifiedThe mainstream metric for evaluating speech naturalness is Mean Opinion Score (MOS), where human listeners rate samples on a scale of 1 to 5.
75%VerifiedOCR accuracy is typically measured using two metrics: Character Error Rate (CER) and Word Error Rate (WER)
75%VerifiedThe formula for Word Error Rate (WER) is (substitutions + deletions + insertions) / total number of words in the reference text, and a lower WER indicates higher transcription quality.
75%VerifiedWER treats all words equally regardless of importance, does not distinguish between errors in key information words and filler words, and cannot measure the readability or semantic accuracy of transcription results.
75%VerifiedThe word error rate formula is WER = (S + D + I) / N, where S is the number of substitution errors, D is the number of deletion errors, I is the number of insertion errors, and N is the total number of words in the reference text.
65%VerifiedVoice drift is a known technical challenge in autoregressive speech generation models, where accumulated errors during step-by-step generation cause the timbre to deviate from the target speaker.
65%VerifiedChildren's shorter vocal tracts result in higher and more variable formant frequencies, which is one of the reasons for the difficulty in children's ASR recognition.
65%UnverifiedCurrent state-of-the-art systems have achieved a WER below 5% on the standard English test set LibriSpeech.
70%UnverifiedEarly mainstream speech recognition solutions relied on Hidden Markov Models (HMM), which had limited capabilities for noise, accent, and multilingual mixed recognition.
70%UnverifiedGemini 3.5 Transcribe can automatically remove false starts and filler words, and identifies self-corrections to retain only the final result.
60%UnverifiedThe formula for relative WER improvement is: (WER_SFT - WER_GRPO) / WER_SFT × 100%.
50%UnverifiedOCR commonly uses character error rate (CER) and word error rate (WER) as evaluation metrics; document AI focuses more on field-level accuracy and the TEDS score for table reconstruction.
50%UnverifiedThe industry commonly uses speaker similarity (such as SECS) and word error rate (WER) as objective evaluation metrics for zero-shot speech cloning.
50%UnverifiedCER is more commonly used for languages without spaces such as Chinese, while WER is better suited to English.
50%UnverifiedCharacter Error Rate (CER) is defined as the edit distance (insertions + deletions + substitutions) divided by the total number of characters in the reference text; Word Error Rate (WER) is analogous but operates on words.
50%UnverifiedThe WER formula is WER = (S + D + I) / N, where S is the number of substitutions, D the number of deletions, I the number of insertions, and N the total number of words in the reference transcription.
50%