Rev AI
Low-cost batch speech-to-text with structured JSON, word timestamps, and speaker labels, but mixed accuracy on hard audio and weak multilingual recall.
Structured and inexpensive, but uneven on hard audio
- You need a batch speech-to-text API that returns JSON with word timestamps, confidence values, and speaker labels.
- You are okay manually auditing speaker attribution and transcript quality on difficult audio.
- You want long-form audio processed asynchronously at a low per-hour cost.
- You need strong multilingual or code-switching accuracy on the configured English tier; Spanish recall was only 8.8%.
Our take
Rev AI consistently returned structured JSON with word-level timestamps, confidence values, and speaker labels, and it was inexpensive at $0.1998 per audio-hour. The catch is accuracy: it hit 9.79% WER on medical jargon, but 28.33% on crosstalk and 25.16% on bilingual code-switching, with only 8.8% Spanish token recall. Compared with the medical-jargon clip, the same configured English tier handled overlap and code-switching much worse, so this looks like a solid batch pipeline rather than a multilingual accuracy winner unless you retest the foreign-language tier.
In-Depth Review
Our detailed analysis of Rev AI — features, performance, and real-world testing.
Feature-by-Feature Breakdown
Asynchronous Batch Transcription▾
Feature tested: Asynchronous Batch Transcription
Result: Partial
Expected behavior: Rev AI submits long-form audio as an async speech-to-text job, then lets you poll for completion and fetch the transcript when the job finishes. The member cards exercised this POST/poll/fetch flow on multiple long-audio benchmark runs, including crosstalk and other long-form cases.
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav (35:43 meeting audio with four-way overlap; 65.39 MB PCM 16 kHz mono) — crosstalk.wav
Observed output: Output artifact (Image): WER 28.33% on the crosstalk sample; the transcript dropped a long middle span, returned 6 speaker labels on audio with 4 participants, and produced 123 insertions and 1,423 deletions against a 7,579-word reference. — 04-transcript-detail.png
Input artifact: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav (35:43 meeting audio with four-way overlap; 65.39 MB PCM 16 kHz mono) — crosstalk.wav
Output artifact: Output artifact (Image): WER 28.33% on the crosstalk sample; the transcript dropped a long middle span, returned 6 speaker labels on audio with 4 participants, and produced 123 insertions and 1,423 deletions against a 7,579-word reference. — 04-transcript-detail.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 (18:44 narration from Gray's Anatomy; 8.58 MB mono MP3) — medical_terms.mp3
Observed output: Output artifact (Image): WER 9.79% on the medical-jargon sample; the transcript missed 'cancellous' and 'trabeculae', but this was the strongest of the three inputs overall. — 04-transcript-detail-2.png
Input artifact: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 (18:44 narration from Gray's Anatomy; 8.58 MB mono MP3) — medical_terms.mp3
Output artifact: Output artifact (Image): WER 9.79% on the medical-jargon sample; the transcript missed 'cancellous' and 'trabeculae', but this was the strongest of the three inputs overall. — 04-transcript-detail-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 (32:18 spontaneous Spanish-English conversation; 22.19 MB stereo MP3) — mix_language.mp3
Observed output: Output artifact (Image): WER 25.16% on the bilingual sample; the transcript dropped Spanish tokens such as 'ahora' and the benchmark measured only 8.8% Spanish token recall, making this effectively weak on code-switching as configured. — 04-transcript-detail-3.png
Input artifact: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 (32:18 spontaneous Spanish-English conversation; 22.19 MB stereo MP3) — mix_language.mp3
Output artifact: Output artifact (Image): WER 25.16% on the bilingual sample; the transcript dropped Spanish tokens such as 'ahora' and the benchmark measured only 8.8% Spanish token recall, making this effectively weak on code-switching as configured. — 04-transcript-detail-3.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: input-1 — Overlapping Speech / Crosstalk (crosstalk.wav, 35:43, 65.4 MB, AMI EN2002a) — crosstalk.wav
Observed output: Output artifact (Image): Raw JSON transcript for crosstalk: the payload shows monologues with word-level elements, timestamps, confidence values, and a footer reporting word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 6249, and distinct_speakers 6. — 02-response-raw.png
Input artifact: Input artifact (Audio file): INPUT: input-1 — Overlapping Speech / Crosstalk (crosstalk.wav, 35:43, 65.4 MB, AMI EN2002a) — crosstalk.wav
Output artifact: Output artifact (Image): Raw JSON transcript for crosstalk: the payload shows monologues with word-level elements, timestamps, confidence values, and a footer reporting word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 6249, and distinct_speakers 6. — 02-response-raw.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: input-2 — Medical Jargon (medical_terms.mp3, 18:44, 8.6 MB, Gray's Anatomy via LibriVox) — medical_terms.mp3
Observed output: Output artifact (Image): Raw JSON transcript for medical jargon: the payload shows word-level elements with timestamps and confidence, and the footer reports word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 2779, and distinct_speakers 1. — 02-response-raw-2.png
Input artifact: Input artifact (Audio file): INPUT: input-2 — Medical Jargon (medical_terms.mp3, 18:44, 8.6 MB, Gray's Anatomy via LibriVox) — medical_terms.mp3
Output artifact: Output artifact (Image): Raw JSON transcript for medical jargon: the payload shows word-level elements with timestamps and confidence, and the footer reports word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 2779, and distinct_speakers 1. — 02-response-raw-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: input-3 — Bilingual Code-Switching (mix_language.mp3, 32:18, 22.2 MB, Bangor Miami herring1) — mix_language.mp3
Observed output: Output artifact (Image): Raw JSON transcript for bilingual code-switching: the payload shows word-level elements with timestamps and confidence, and the footer reports word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 5855, and distinct_speakers 3. — 02-response-raw-3.png
Input artifact: Input artifact (Audio file): INPUT: input-3 — Bilingual Code-Switching (mix_language.mp3, 32:18, 22.2 MB, Bangor Miami herring1) — mix_language.mp3
Output artifact: Output artifact (Image): Raw JSON transcript for bilingual code-switching: the payload shows word-level elements with timestamps and confidence, and the footer reports word_timestamps yes, confidence yes, speaker_labels yes, timed_tokens 5855, and distinct_speakers 3. — 02-response-raw-3.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav (65.39 MB, 2142.709s, four-way overlapping meeting audio) — crosstalk.wav
Observed output: Output artifact (Image): Scored execution trace for the crosstalk run: the async pipeline completed end to end, the run was scored at 28.33% WER, latency was 102.75s, cost was $0.11892, and the trace records 6 speaker labels against a 4-participant reference. — 07-automation-trace.png
Input artifact: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav (65.39 MB, 2142.709s, four-way overlapping meeting audio) — crosstalk.wav
Output artifact: Output artifact (Image): Scored execution trace for the crosstalk run: the async pipeline completed end to end, the run was scored at 28.33% WER, latency was 102.75s, cost was $0.11892, and the trace records 6 speaker labels against a 4-participant reference. — 07-automation-trace.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 (8.58 MB, 1123.944s, single-speaker anatomy narration) — medical_terms.mp3
Observed output: Output artifact (Image): Scored execution trace for the medical-jargon run: the async pipeline completed end to end, the run was scored at 9.79% WER, latency was 78.32s, cost was $0.06238, and the trace records the benchmark truth file used for scoring. — 07-automation-trace-2.png
Input artifact: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 (8.58 MB, 1123.944s, single-speaker anatomy narration) — medical_terms.mp3
Output artifact: Output artifact (Image): Scored execution trace for the medical-jargon run: the async pipeline completed end to end, the run was scored at 9.79% WER, latency was 78.32s, cost was $0.06238, and the trace records the benchmark truth file used for scoring. — 07-automation-trace-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 (22.19 MB, 1938.495s, Spanish-English conversation) — mix_language.mp3
Observed output: Output artifact (Image): Scored execution trace for the bilingual code-switching run: the async pipeline completed end to end, the run was scored at 25.16% WER, latency was 99.79s, cost was $0.10759, and the trace records the benchmark truth file used for scoring. — 07-automation-trace-3.png
Input artifact: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 (22.19 MB, 1938.495s, Spanish-English conversation) — mix_language.mp3
Output artifact: Output artifact (Image): Scored execution trace for the bilingual code-switching run: the async pipeline completed end to end, the run was scored at 25.16% WER, latency was 99.79s, cost was $0.10759, and the trace records the benchmark truth file used for scoring. — 07-automation-trace-3.png
What changed: Audio file transformed into Image
Why it matters / Conclusion: The async integration pattern worked cleanly across all three files, returning transcript JSON each time; the benchmark's quality differences came from the audio difficulty, not from request handling failures.
Rev AI submits long-form audio as an async speech-to-text job, then lets you poll for completion and fetch the transcript when the job finishes. The member cards exercised this POST/poll/fetch flow on multiple long-audio benchmark runs, including crosstalk and other long-form cases.
Structured Transcript Output▾
Feature tested: Structured Transcript Output
Result: Partial
Expected behavior: Rev AI returns transcription results as structured JSON with rich metadata such as word-level timestamps, confidence values, punctuation tokens, monologues, and speaker-label fields. The member cards show this payload structure consistently across runs, supporting downstream transcript tooling.
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav — crosstalk.wav
Observed output: Output artifact (Image): Raw API response for the crosstalk run showing monologues with word-level tokens, timestamps, confidence values, and detected developer features such as word_timestamps, confidence, and speaker_labels. — 02-response-raw.png
Input artifact: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav — crosstalk.wav
Output artifact: Output artifact (Image): Raw API response for the crosstalk run showing monologues with word-level tokens, timestamps, confidence values, and detected developer features such as word_timestamps, confidence, and speaker_labels. — 02-response-raw.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 — medical_terms.mp3
Observed output: Output artifact (Image): Raw API response for the medical-jargon run showing word-level tokens, timestamps, confidence values, and detected developer features including timed word tokens and speaker label fields. — 02-response-raw-2.png
Input artifact: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 — medical_terms.mp3
Output artifact: Output artifact (Image): Raw API response for the medical-jargon run showing word-level tokens, timestamps, confidence values, and detected developer features including timed word tokens and speaker label fields. — 02-response-raw-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Observed output: Output artifact (Image): Raw API response for the bilingual code-switching run showing monologues with word-level tokens, timestamps, confidence values, and detected developer features including timed tokens and speaker labels. — 02-response-raw-3.png
Input artifact: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Output artifact: Output artifact (Image): Raw API response for the bilingual code-switching run showing monologues with word-level tokens, timestamps, confidence values, and detected developer features including timed tokens and speaker labels. — 02-response-raw-3.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav — crosstalk.wav
Observed output: Output artifact (Image): Transcript detail for the crosstalk run showing the largest divergence, WER 28.33%, and diarization detected 6 labels even though the reference has 4 participants. — 04-transcript-detail.png
Input artifact: Input artifact (Audio file): Overlapping Speech / Crosstalk — crosstalk.wav — crosstalk.wav
Output artifact: Output artifact (Image): Transcript detail for the crosstalk run showing the largest divergence, WER 28.33%, and diarization detected 6 labels even though the reference has 4 participants. — 04-transcript-detail.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 — medical_terms.mp3
Observed output: Output artifact (Image): Run-metrics summary for the medical-jargon run showing diarization-related payload fields alongside WER 9.79%, latency 78.32s, RTF 0.06968, and estimated cost $0.06238. — 03-terminal-metrics-2.png
Input artifact: Input artifact (Audio file): Medical Jargon — medical_terms.mp3 — medical_terms.mp3
Output artifact: Output artifact (Image): Run-metrics summary for the medical-jargon run showing diarization-related payload fields alongside WER 9.79%, latency 78.32s, RTF 0.06968, and estimated cost $0.06238. — 03-terminal-metrics-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Observed output: Output artifact (Image): Run-metrics summary for the bilingual code-switching run showing diarization-related payload fields alongside WER 25.16%, latency 99.79s, RTF 0.05148, and estimated cost $0.10759. — 03-terminal-metrics-3.png
Input artifact: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Output artifact: Output artifact (Image): Run-metrics summary for the bilingual code-switching run showing diarization-related payload fields alongside WER 25.16%, latency 99.79s, RTF 0.05148, and estimated cost $0.10759. — 03-terminal-metrics-3.png
What changed: Audio file transformed into Image
Why it matters / Conclusion: This structure was consistent on every input; however, the report does not separately score timestamp alignment, so the evidence here is about payload richness rather than timing accuracy.
Rev AI returns transcription results as structured JSON with rich metadata such as word-level timestamps, confidence values, punctuation tokens, monologues, and speaker-label fields. The member cards show this payload structure consistently across runs, supporting downstream transcript tooling.
Speaker DiarizationDetected, but overlap was over-segmented▾
Feature tested: Speaker Diarization
Result: Partial
Verdict: Detected, but overlap was over-segmented
Expected behavior: Rev AI can emit speaker labels in transcript output so multi-speaker audio can be separated by speaker. The evidence includes overlapping-meeting and crosstalk samples where labels were produced, though segmentation quality varied.
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — crosstalk.wav, 35:43, 65.39 MB, AMI EN2002a. — crosstalk.wav
Observed output: Output artifact (Image): Transcript detail for crosstalk: the largest divergence is shown alongside WER 28.33%, and the note reports diarization detected 6 labels while the ground truth had 4 participants. — 04-transcript-detail.png
Input artifact: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — crosstalk.wav, 35:43, 65.39 MB, AMI EN2002a. — crosstalk.wav
Output artifact: Output artifact (Image): Transcript detail for crosstalk: the largest divergence is shown alongside WER 28.33%, and the note reports diarization detected 6 labels while the ground truth had 4 participants. — 04-transcript-detail.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: Medical Jargon — medical_terms.mp3, 18:44, 8.58 MB, Gray's Anatomy. — medical_terms.mp3
Observed output: Output artifact (Image): Raw response preview for medical jargon: the payload includes speaker labels and reports 1 distinct speaker in the detected output. — 02-response-raw-2.png
Input artifact: Input artifact (Audio file): INPUT: Medical Jargon — medical_terms.mp3, 18:44, 8.58 MB, Gray's Anatomy. — medical_terms.mp3
Output artifact: Output artifact (Image): Raw response preview for medical jargon: the payload includes speaker labels and reports 1 distinct speaker in the detected output. — 02-response-raw-2.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: Bilingual Code-Switching — mix_language.mp3, 32:18, 22.19 MB, Bangor Miami herring1. — mix_language.mp3
Observed output: Output artifact (Image): Raw response preview for bilingual code-switching: the payload includes speaker labels and reports 3 distinct speakers in the detected output. — 02-response-raw-3.png
Input artifact: Input artifact (Audio file): INPUT: Bilingual Code-Switching — mix_language.mp3, 32:18, 22.19 MB, Bangor Miami herring1. — mix_language.mp3
Output artifact: Output artifact (Image): Raw response preview for bilingual code-switching: the payload includes speaker labels and reports 3 distinct speakers in the detected output. — 02-response-raw-3.png
What changed: Audio file transformed into Image
Why it matters / Conclusion: It does expose speaker labels, but the overlap case was over-segmented and the benchmark did not measure whether the labels were attributed to the right speakers.
Rev AI can emit speaker labels in transcript output so multi-speaker audio can be separated by speaker. The evidence includes overlapping-meeting and crosstalk samples where labels were produced, though segmentation quality varied.
Code-Switching Transcription▾
Feature tested: Code-Switching Transcription
Result: Failed
Expected behavior: Rev AI attempts transcription on mixed-language audio, including a Spanish-English code-switching conversation. The benchmarked runs showed weak Spanish recall in the tested configuration, but both cards are exercising the same bilingual transcription capability.
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Observed output: Output artifact (Image): Run-metrics summary for the bilingual code-switching run showing WER 25.16%, Spanish recall 8.8%, and the measured latency, RTF, and cost for the configured English Reverb model. — 03-terminal-metrics-3.png
Input artifact: Input artifact (Audio file): Bilingual Code-Switching — mix_language.mp3 — mix_language.mp3
Output artifact: Output artifact (Image): Run-metrics summary for the bilingual code-switching run showing WER 25.16%, Spanish recall 8.8%, and the measured latency, RTF, and cost for the configured English Reverb model. — 03-terminal-metrics-3.png
What changed: Audio file transformed into Image
Test case: Audio file → Image
Input type: Audio file
Input used: Input artifact (Audio file): INPUT: Bilingual Code-Switching — mix_language.mp3, 32:18, 22.19 MB, Bangor Miami herring1. — mix_language.mp3
Observed output: Output artifact (Image): Transcript detail for bilingual code-switching: the error site shows Spanish 'mi entonces ahora' dropped from the output, with WER 25.16% and Spanish recall 8.8%. — 04-transcript-detail-3.png
Input artifact: Input artifact (Audio file): INPUT: Bilingual Code-Switching — mix_language.mp3, 32:18, 22.19 MB, Bangor Miami herring1. — mix_language.mp3
Output artifact: Output artifact (Image): Transcript detail for bilingual code-switching: the error site shows Spanish 'mi entonces ahora' dropped from the output, with WER 25.16% and Spanish recall 8.8%. — 04-transcript-detail-3.png
What changed: Audio file transformed into Image
Why it matters / Conclusion: The configured English tier was effectively weak on Spanish code-switching; a fair multilingual retest would need the separate Foreign Language tier.
Rev AI attempts transcription on mixed-language audio, including a Spanish-English code-switching conversation. The benchmarked runs showed weak Spanish recall in the tested configuration, but both cards are exercising the same bilingual transcription capability.
How it scored on the research's own criteria
The 4 evaluation dimensions from our hands-on research on Rev AI, each judged from recorded runs on 3 test inputs — the same verdicts the ranking page ranks on.
held up partial failed not exercised by this input
| Criterion | Verdict | What the runs showed | Per input | Proof |
|---|---|---|---|---|
| Output quality | Weak2/5 | One narration run is usable but still leaves notable lexical gaps, while the crosstalk and bilingual runs lose too many words and even split speakers incorrectly, so accuracy is not dependable in harder audio. | open proof ↗ | |
| Automation level | Strong5/5 | Each run went from submission to transcript fetch without manual intervention, so the workflow is genuinely hands-off even though the step count follows the vendor's documented protocol. | — | |
| Export | Strong5/5 | Every run returned the transcript plus timing, confidence, and speaker labeling, so downstream use gets the full machine-readable package rather than a bare text dump. | — | |
| Input handling | Strong5/5 | All three runs took the audio, finished, and stayed well under real time with no rejection, so the service handled ingestion reliably rather than merely accepting small or easy files. | open proof ↗ |
Verdicts come verbatim from the study's recorded observations, never re-derived at render; a criterion with no recorded run shows Not exercised — this section cannot invent a score.
Official pricing
The benchmark used Reverb Transcription (English) at $0.20 per audio-hour.
Free tier includes 5 audio-hours of credits. Multilingual work requires the separate Reverb Foreign Language tier; language identification and other add-ons are billed separately.
Banner Preview
How the embed badge will look on your site
Embed HTML
Copy this code to your website source
Quick Integration Guide
- 1Copy the HTML code block above.
- 2Paste it into your site's HTML or CMS editor.
- 3Banner appears instantly on your page.
- 4Links back to your tool profile here.
Similar Tools
Discover more AI tools like Rev AI to enhance your workflow.
Comments (0)
Need a custom AI solution for this use case?
If you are looking to build a custom speech-to-text, audio transcription, or diarization system for your business or internal workflow, email us at contact@futuresmart.ai.
Found something inaccurate or missing? We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at collaborate@aidemos.com.