← All models

Azure Speech-to-Text

Audio Available Full comparison ↗

microsoft-azure/speech-stt · by Microsoft · end-to-end-asr

Pricing — 1 offering(s)

Standard transcription

  • $0.0167 / minute Current 2017-06-01 → present
    USD $1.00/hour pay-as-you-go. Free tier: 5 hours/month. SCO-608 re-verification: unchanged ("Standard…

    USD $1.00/hour pay-as-you-go. Free tier: 5 hours/month. SCO-608 re-verification: unchanged ("Standard Transcription Real-time Transcription: $1 per hour" on the live page). Source URL updated — Microsoft rebranded/restructured the product as "Azure Speech in Foundry Tools" and the old cognitive-services/speech-services/ path now redirects here rather than being canonical. Flag for a follow-up, not fixed here: the live page also lists a separate, cheaper "Batch Transcription: $0.18 per hour" rate not currently tracked as its own tier on this entry — a coverage gap, not a staleness issue.

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Azure Speech-to-Text fits into a cost-aware routing setup

See how →

Capability profile

How Azure Speech-to-Text rates across core capability dimensions, with the task-level evidence behind each rating.

transcription accuracy Strong

High accuracy on clean audio and well-structured speech. Custom Speech model training further improves domain-specific accuracy. Strong on formal/business speech.

language support Strong

100+ languages and locales. Multilingual transcription (multiple languages in the same audio) supported.

speaker diarisation Strong

Built-in speaker diarisation. Conversation transcription endpoint is optimised for multi-speaker scenarios (meetings, calls).

real time streaming Strong

WebSocket streaming via the Speech SDK. Stable and production-grade for enterprise deployments.

custom vocabulary Strong

Custom Speech training pipeline: upload domain audio and transcripts to fine-tune models. Phrase lists available for lightweight customisation.

Benchmarks

Benchmark Score Config Source
WER (published, Microsoft Word Error Rate report) — Microsoft publishes periodic WER comparisons in their Azure blog. Numbers are vendor-reported. No independent third-party comparison on LibriSpeech available. Checked 2026-08-19 (SCO-462 sweep): the Artificial Analysis Speech-to-Text leaderboard (artificialanalysis.ai/speech-to-text) lists "MAI-Transcribe-1.5" (2.4% WER) and "MAI-Transcribe-1" (2.6% WER) under Microsoft/Azure — NOT cited as this model's score. "MAI" (Microsoft AI) is a distinct in-house model family: confirmed 2026-08-29 (SCO-466) that MAI-Transcribe-1/1.5 is NOT the Azure Cognitive Services Speech-to-Text API this entry prices — it is a separate "LLM Speech" SKU ($0.36/hour, Microsoft Foundry). Those WER figures stay uncited here; MAI-Transcribe now has its own entry (microsoft-azure/mai-transcribe, SCO-507). No independent third-party WER for this specific Cognitive Services STT product. —

Operator guidance

Best for enterprise teams on Azure needing compliance, custom model training, and Microsoft ecosystem integration (Teams, Dynamics). At $1.00/hour ($0.0167/min) it is the most expensive major STT option but provides the strongest enterprise governance story. For cost, Universal-2 ($0.15/hr) is 6× cheaper. For streaming latency, Deepgram Nova-3 is faster.

Use cases

Limitations

Citations