AWS icon
audio-speech

AWS

Batch speech-to-text with timestamps and speaker labels, but weak on multilingual audio.

Visit AWS
Batch STTWord timestampsSpeaker labelsCode-switching tested
TL;DR — our verdictUpdated August 2026 · 9 test artifacts

Solid batch API, but not a safe multilingual pick as configured

Where it wins
  • you need a batch transcription API that returns word-level timestamps, confidence, and speaker labels
  • you can stage audio in S3 and run an async job/poll workflow
  • you are transcribing mostly English or technical jargon and can tolerate weaker performance on harder audio
Main limitation
  • you need reliable multilingual or code-switching transcription
Pricing (verified plans)
Free tier $0Standard Batch $0.0001000/secStandard Streaming $0.0001667/sec
Strongest test artifacts
Batch Speech-to-Text Transcription →Word-Level Metadata Export →

Our take

AWS Transcribe returned a complete developer payload on every scored run and handled the medical jargon clip well, but the overlap sample lost a lot of content and the Spanish-English sample showed weak Spanish recovery. In this benchmark it looks like a dependable batch transcription API for English-centric workflows, not a strong choice for multilingual or code-switching audio without further tuning.

Download video: Demo videoWatch video (streaming): Demo video
Tutorial recording of the AWS Transcribe benchmark workflow.

In-Depth Review

Our detailed analysis of AWS — features, performance, and real-world testing.

AD
AI Demos Team
Expert Reviewer
Verified Review

Feature-by-Feature Breakdown

Batch Speech-to-Text Transcription
Mixed
▾
Test Summary
Feature tested: Batch Speech-to-Text Transcription
Result: Failed — Mixed

Feature tested: Batch Speech-to-Text Transcription

Result: Failed

Verdict: Mixed

Expected behavior: Runs asynchronous batch transcription jobs on long-form audio staged in AWS and returns full transcripts. It was exercised on overlapping crosstalk, medical-jargon, and bilingual code-switching audio, with results varying by difficulty.

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, four-way overlapping meeting audio). — crosstalk.wav

Observed output: Output artifact (Image): WER 33.88% on the overlapping-speech run, with 5460 words returned against a 7579-word reference and the largest divergence showing a substantial mismatch at the highlighted error site. — 04-transcript-detail.png

Input artifact: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — crosstalk.wav (35:43, 65.39 MB, four-way overlapping meeting audio). — crosstalk.wav

Output artifact: Output artifact (Image): WER 33.88% on the overlapping-speech run, with 5460 words returned against a 7579-word reference and the largest divergence showing a substantial mismatch at the highlighted error site. — 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, clean narration with anatomical terminology). — medical_terms.mp3

Observed output: Output artifact (Image): WER 3.63% on the medical-jargon run, with the highlighted error site showing 'cancerous' where the reference said 'cancellous' and jargon recall at 77.8%. — 04-transcript-detail-2.png

Input artifact: Input artifact (Audio file): INPUT: Medical Jargon — medical_terms.mp3 (18:44, 8.58 MB, clean narration with anatomical terminology). — medical_terms.mp3

Output artifact: Output artifact (Image): WER 3.63% on the medical-jargon run, with the highlighted error site showing 'cancerous' where the reference said 'cancellous' and jargon recall at 77.8%. — 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): INPUT: Bilingual Code-switching — mix_language.mp3 (Spanish-English conversation). — mix_language.mp3

Observed output: Output artifact (Image): The transcript-detail view flags a dropped Spanish token at the error site and shows the mixed-language divergence on the code-switching sample. — 04-transcript-detail-3.png

Input artifact: Input artifact (Audio file): INPUT: Bilingual Code-switching — mix_language.mp3 (Spanish-English conversation). — mix_language.mp3

Output artifact: Output artifact (Image): The transcript-detail view flags a dropped Spanish token at the error site and shows the mixed-language divergence on the code-switching sample. — 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: Overlapping Speech / Crosstalk — verify the end-to-end batch job flow. — crosstalk.wav

Observed output: Output artifact (Image): Execution trace for crosstalk: multi-stage vendor protocol, no measured per-call count, 196.83s wall-clock latency, and 5,460 returned words from a 7,579-word reference. — 07-automation-trace.png

Input artifact: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — verify the end-to-end batch job flow. — crosstalk.wav

Output artifact: Output artifact (Image): Execution trace for crosstalk: multi-stage vendor protocol, no measured per-call count, 196.83s wall-clock latency, and 5,460 returned words from a 7,579-word 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): INPUT: Medical Jargon — verify the end-to-end batch job flow. — medical_terms.mp3

Observed output: Output artifact (Image): Execution trace for medical jargon: multi-stage vendor protocol, no measured per-call count, 96.835s latency, and 2,728 returned words. — 07-automation-trace-2.png

Input artifact: Input artifact (Audio file): INPUT: Medical Jargon — verify the end-to-end batch job flow. — medical_terms.mp3

Output artifact: Output artifact (Image): Execution trace for medical jargon: multi-stage vendor protocol, no measured per-call count, 96.835s latency, and 2,728 returned words. — 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): INPUT: Bilingual Code-Switching — verify the end-to-end batch job flow. — mix_language.mp3

Observed output: Output artifact (Image): Execution trace for bilingual code-switching: multi-stage vendor protocol, no measured per-call count, 172.18s latency, and 6,017 returned words. — 07-automation-trace-3.png

Input artifact: Input artifact (Audio file): INPUT: Bilingual Code-Switching — verify the end-to-end batch job flow. — mix_language.mp3

Output artifact: Output artifact (Image): Execution trace for bilingual code-switching: multi-stage vendor protocol, no measured per-call count, 172.18s latency, and 6,017 returned words. — 07-automation-trace-3.png

What changed: Audio file transformed into Image

Why it matters / Conclusion: Best on the medical-jargon clip, acceptable on the code-switching clip in headline WER, and clearly weakest on crosstalk because of heavy deletions.

Runs asynchronous batch transcription jobs on long-form audio staged in AWS and returns full transcripts. It was exercised on overlapping crosstalk, medical-jargon, and bilingual code-switching audio, with results varying by difficulty.

audio
0:00 / 0:00
Loading audio...
INPUT: Overlapping Speech / Crosstalk — crosstalk.wav (35:43, 65.39 MB, four-way overlapping meeting audio).
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: WER 33.88% on the overlapping-speech run, with 5460 words returned against a 7579-word reference and the largest divergence showing a substantial mismatch at the highlighted error site., 04-transcript-detail.png
WER 33.88% on the overlapping-speech run, with 5460 words returned against a 7579-word reference and the largest divergence showing a substantial mismatch at the highlighted error site.
audio
0:00 / 0:00
Loading audio...
INPUT: Medical Jargon — medical_terms.mp3 (18:44, 8.58 MB, clean narration with anatomical terminology).
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: WER 3.63% on the medical-jargon run, with the highlighted error site showing 'cancerous' where the reference said 'cancellous' and jargon recall at 77.8%., 04-transcript-detail-2.png
WER 3.63% on the medical-jargon run, with the highlighted error site showing 'cancerous' where the reference said 'cancellous' and jargon recall at 77.8%.
audio
0:00 / 0:00
Loading audio...
INPUT: Bilingual Code-switching — mix_language.mp3 (Spanish-English conversation).
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: The transcript-detail view flags a dropped Spanish token at the error site and shows the mixed-language divergence on the code-switching sample., 04-transcript-detail-3.png
The transcript-detail view flags a dropped Spanish token at the error site and shows the mixed-language divergence on the code-switching sample.
audio
0:00 / 0:00
Loading audio...
INPUT: Overlapping Speech / Crosstalk — verify the end-to-end batch job flow.
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: Execution trace for crosstalk: multi-stage vendor protocol, no measured per-call count, 196.83s wall-clock latency, and 5,460 returned words from a 7,579-word reference., 07-automation-trace.png
Execution trace for crosstalk: multi-stage vendor protocol, no measured per-call count, 196.83s wall-clock latency, and 5,460 returned words from a 7,579-word reference.
audio
0:00 / 0:00
Loading audio...
INPUT: Medical Jargon — verify the end-to-end batch job flow.
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: Execution trace for medical jargon: multi-stage vendor protocol, no measured per-call count, 96.835s latency, and 2,728 returned words., 07-automation-trace-2.png
Execution trace for medical jargon: multi-stage vendor protocol, no measured per-call count, 96.835s latency, and 2,728 returned words.
audio
0:00 / 0:00
Loading audio...
INPUT: Bilingual Code-Switching — verify the end-to-end batch job flow.
↓
OUTPUT
Output artifact for "Batch Speech-to-Text Transcription" test: Execution trace for bilingual code-switching: multi-stage vendor protocol, no measured per-call count, 172.18s latency, and 6,017 returned words., 07-automation-trace-3.png
Execution trace for bilingual code-switching: multi-stage vendor protocol, no measured per-call count, 172.18s latency, and 6,017 returned words.
Bottom Line
Best on the medical-jargon clip, acceptable on the code-switching clip in headline WER, and clearly weakest on crosstalk because of heavy deletions.
From our researchearlier researchTranscribe Audio Accurately — Speech-to-Text Engine Benchmark
Word-Level Metadata Export
Strong
▾
Test Summary
Feature tested: Word-Level Metadata Export
Result: Passed — Strong

Feature tested: Word-Level Metadata Export

Result: Passed

Verdict: Strong

Expected behavior: Returns developer-facing JSON with word-level timings, confidence values, and speaker labels. Across the scored runs, the payload consistently exposed these fields and supported downstream tooling.

Test case: Audio file → Image

Input type: Audio file

Input used: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — crosstalk.wav (batch transcription run). — crosstalk.wav

Observed output: Output artifact (Image): Raw response for the crosstalk run shows word timestamps, confidence, speaker labels, 5913 linked tokens, and 4 distinct speakers. — 02-response-raw.png

Input artifact: Input artifact (Audio file): INPUT: Overlapping Speech / Crosstalk — crosstalk.wav (batch transcription run). — crosstalk.wav

Output artifact: Output artifact (Image): Raw response for the crosstalk run shows word timestamps, confidence, speaker labels, 5913 linked tokens, and 4 distinct speakers. — 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: Medical Jargon — medical_terms.mp3 (batch transcription run). — medical_terms.mp3

Observed output: Output artifact (Image): Raw response for the medical-jargon run again shows word timestamps, confidence, and speaker labels, with 2829 linked tokens and 4 distinct speakers. — 02-response-raw-2.png

Input artifact: Input artifact (Audio file): INPUT: Medical Jargon — medical_terms.mp3 (batch transcription run). — medical_terms.mp3

Output artifact: Output artifact (Image): Raw response for the medical-jargon run again shows word timestamps, confidence, and speaker labels, with 2829 linked tokens and 4 distinct speakers. — 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 (batch transcription run). — mix_language.mp3

Observed output: Output artifact (Image): Raw response for the bilingual run shows word times, confidence, and speaker labels, with 6311 labeled tokens and 3 distinct speakers. — 02-response-raw-3.png

Input artifact: Input artifact (Audio file): INPUT: Bilingual Code-switching — mix_language.mp3 (batch transcription run). — mix_language.mp3

Output artifact: Output artifact (Image): Raw response for the bilingual run shows word times, confidence, and speaker labels, with 6311 labeled tokens and 3 distinct speakers. — 02-response-raw-3.png

What changed: Audio file transformed into Image

Why it matters / Conclusion: Consistent across every scored run and useful for downstream tooling, but the report only verifies that labels exist and counts them — it does not verify diarization attribution.

Returns developer-facing JSON with word-level timings, confidence values, and speaker labels. Across the scored runs, the payload consistently exposed these fields and supported downstream tooling.

audio
0:00 / 0:00
Loading audio...
INPUT: Overlapping Speech / Crosstalk — crosstalk.wav (batch transcription run).
↓
OUTPUT
Output artifact for "Word-Level Metadata Export" test: Raw response for the crosstalk run shows word timestamps, confidence, speaker labels, 5913 linked tokens, and 4 distinct speakers., 02-response-raw.png
Raw response for the crosstalk run shows word timestamps, confidence, speaker labels, 5913 linked tokens, and 4 distinct speakers.
audio
0:00 / 0:00
Loading audio...
INPUT: Medical Jargon — medical_terms.mp3 (batch transcription run).
↓
OUTPUT
Output artifact for "Word-Level Metadata Export" test: Raw response for the medical-jargon run again shows word timestamps, confidence, and speaker labels, with 2829 linked tokens and 4 distinct speakers., 02-response-raw-2.png
Raw response for the medical-jargon run again shows word timestamps, confidence, and speaker labels, with 2829 linked tokens and 4 distinct speakers.
audio
0:00 / 0:00
Loading audio...
INPUT: Bilingual Code-switching — mix_language.mp3 (batch transcription run).
↓
OUTPUT
Output artifact for "Word-Level Metadata Export" test: Raw response for the bilingual run shows word times, confidence, and speaker labels, with 6311 labeled tokens and 3 distinct speakers., 02-response-raw-3.png
Raw response for the bilingual run shows word times, confidence, and speaker labels, with 6311 labeled tokens and 3 distinct speakers.
Bottom Line
Consistent across every scored run and useful for downstream tooling, but the report only verifies that labels exist and counts them — it does not verify diarization attribution.
From our researchearlier researchTranscribe Audio Accurately — Speech-to-Text Engine Benchmark

How it scored on the research's own criteria

The 4 evaluation dimensions from our hands-on research on AWS , 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

CriterionVerdictWhat the runs showedPer inputProof
Automation levelStrong4/5The job runs all the way through without someone steering each step, which is strong automation. It does not earn the maximum because the trace never measures the individual API hops, so the exact level of orchestration is partly inferred from the documented flow rather than observed directly.open proof ↗
ExportStrong5/5Every run returned a full structured payload with transcript text plus timing, confidence, and speaker labeling. Because that structure is present on all tested inputs, this is a clear maximum score.open proof ↗
Input handlingStrong5/5It handled every uploaded file cleanly and finished all three runs while reporting latency, real-time factor, and cost each time. That is consistent end-to-end acceptance, not a partial or flaky run, so this belongs at the top of the scale.open proof ↗
Output qualityMixed3/5One clean narration run was excellent, but the two messier speech conditions fell off sharply, especially overlap and Spanish switching. That makes the tool useful on straightforward audio but inconsistent overall rather than reliably accurate across the set.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.

AWS Transcribe pricing

Standard batch and streaming are flat-rate; the benchmarked runs here used batch transcription.

Free tier
$0
60 audio minutes/month for 12 months; unused minutes do not roll over.
TESTED
Standard Batch
$0.0001000/sec ($0.006/min)
Flat rate, no volume tiers; billed in 1-second increments; up to two channels included.
Standard Streaming
$0.0001667/sec ($0.010002/min)
Flat rate, no volume tiers.

AWS's own pricing feed shows standard transcription uses a single rate dimension with no volume tiers; add-ons such as PII redaction and custom language models are billed separately. The benchmarked cost figures in this report are list price multiplied by measured duration.

✓ Use This If
●you need a batch transcription API that returns word-level timestamps, confidence, and speaker labels
●you can stage audio in S3 and run an async job/poll workflow
●you are transcribing mostly English or technical jargon and can tolerate weaker performance on harder audio
✕ Skip This If
●you need reliable multilingual or code-switching transcription
●you need verified diarization attribution rather than just detected speaker labels
●you need measured streaming latency from this benchmark
●you need direct local-file transcription without S3 staging
audio-speechaudio-to-textspeechOther
The report sources AWS's own pricing feed and lists standard batch transcription at $0.006/minute ($0.36/hour), standard streaming at about $0.010002/minute, and a free tier of 60 audio minutes per month for 12 months.
Yes. The report says audio must be staged in S3 first before the batch transcription job runs.
Yes. All three tested runs exposed word timestamps and confidence fields in the raw JSON response.
Yes. Speaker labels were detected on every scored run, with 4 distinct speakers on the crosstalk and medical-jargon inputs and 3 distinct speakers on the bilingual input. The report does not measure whether those labels were attributed correctly.
Poorly. On the crosstalk run it scored 33.88% WER, returned 5460 words against a 7579-word reference, and dropped 2162 words.
Weakly as configured. On the Spanish-English sample the benchmark summary recorded only 30.0% Spanish token recall, so the report recommends re-testing with explicit multilingual settings before trusting it for bilingual audio.

Banner Preview

How the embed badge will look on your site

AWS featured on AI Demos

Embed HTML

Copy this code to your website source

<a target="_blank" href="https://aidemos.com/tools/aws-transcribe?utm_source=aws-transcribe_embed" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> <img src="https://aidemos-website-images.s3.amazonaws.com/featured.png" alt="AWS | Featured on AI Demos" style="width: 250px; height: 80px; border-radius:4px;" width="250" height="80"> </a>

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

Similar Tools

Discover more AI tools like AWS to enhance your workflow.

🤖
Gladia Solaria
AI Tool
🤖
Deepgram Nova-3
Batch speech-to-text with rich metadata, but weak on crosstalk and code-switching.
AI Tool
🤖
AssemblyAI Universal
Fast batch STT with strong metadata and mixed-language performance, but overlap-heavy meetings can drop too many words.
AI Tool
🤖
Speechmatics Ursa/Enhanced
AI Tool
🤖
OpenAI Whisper
AI Tool
🤖
Google Cloud Speech-to-Text
AI Tool
🤖
Microsoft Azure Speech Services
AI Tool
🤖
ElevenLabs Scribe
Fast batch transcription with rich metadata, strongest on jargon and weaker on overlap and code-switching.
AI Tool
🤖
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.
AI Tool
🤖
GroqCloud Whisper Large-v3
Fast batch transcription for under-cap audio, with strong jargon handling but a hard 25 MB ceiling.
AI Tool
🤖
Speechmatics Ursa Enhanced
AI Tool
🤖
OpenAI Whisper-1
AI Tool
🤖
Google Cloud Speech-to-Text Chirp 2
AI Tool

Comments (0)

Please Log in to join the discussion.

Built by FutureSmart AI — the team behind AI Demos

Need a custom AI solution for this use case?

If you are looking to build a custom speech-to-text transcription, timestamping, or speaker diarization workflow for your business or internal workflow, email us at contact@futuresmart.ai.

Get a custom build

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.

Back to Top