Tool profile
Azure AI Speech: cost, fit and limits
Azure AI Speech is a Microsoft cloud service that provides speech to text through real-time, fast, batch, custom and LLM-based transcription tiers, billed per audio hour sent. It is part of the Azure platform and includes a free F0 tier.
What to know first
Azure AI Speech lists real-time, fast, batch, MAI Transcribe and LLM Speech Transcription under Standard, plus Custom transcription with hosted endpoints. Billing is per audio hour in second increments. The free F0 tier gives 5 audio hours of real-time transcription a month, and commitment tiers exist at 2,000, 10,000 and 50,000 hours. Per-hour rates did not render, so no price is recorded here.
Azure AI Speech at a glance
| Maker | Microsoft, United States |
|---|---|
| Free plan | Yes |
| Entry price | Quote only |
| API access | Yes |
| Category | Best AI Speech to Text APIs in 2026, rank 7 of 12 |
| Website | azure.microsoft.com/products/ai-services/ai-speech |
Who Azure AI Speech is best for
Microsoft-centered organizations
Billing, access control and governance stay in Azure, which helps teams already standardized on Microsoft cloud tooling. It also suits teams that want one contract and one security review for speech and other AI services.
Teams needing custom acoustic or language models
Custom transcription with endpoint hosting per model suits specialized vocabularies, provided the team keeps the endpoint in use and accepts per-hour hosting charges.
Large steady workloads
Monthly commitment tiers of 2,000, 10,000 and 50,000 hours give predictable volume pricing for contact centers and media archives. The tiers matter most when volume is steady enough that a monthly commitment actually gets used up.
Azure AI Speech plans and pricing
| Plan | Price | What it includes |
|---|---|---|
| Free (F0) | Free | 5 audio hours of real-time transcription per month, shared with Custom. Batch is not supported on the Standard free tier. |
| Standard | Per-hour rate not rendered on the opened page | Real-time, Fast, MAI Transcribe-2 and -1.5, Batch, and LLM Speech Transcription tiers are listed. Add-ons such as diarization are billed per hour on top of these. |
| Custom | Per-hour rate not rendered on the opened page | Custom real-time and batch transcription, endpoint hosting per model per hour, and commitment tiers for 2,000, 10,000 and 50,000 hours a month. |
Checked October 2026.
What Azure AI Speech does
- Multiple transcription tiers
- Real-time, Fast, Batch and LLM Speech Transcription are listed separately, so a team can pick latency and cost per workload. Fast transcription and LLM transcription are listed apart from batch, which suggests different latency and cost profiles.
- Named Microsoft models
- The page lists MAI Transcribe-2 and MAI Transcribe-1.5 as Standard tiers, alongside an LLM-based transcription tier. The page does not explain how these models differ in quality, so buyers need their own tests.
- Per-feature add-ons
- Continuous language identification, diarization and pronunciation assessment are priced per hour per feature, so each one adds to the base transcription cost. A diarized, language-identified transcript therefore costs more than the plain transcription rate, by an amount that depends on the add-ons chosen.
- Custom endpoints
- Custom transcription models are hosted at an hourly endpoint charge per model, in addition to the transcription rate itself. This is a standing cost for every model deployed, whether or not audio is flowing through it.
How Azure AI Speech handles your data
The Azure AI Speech pricing page does not state a retention period, a training policy, regional hosting or certifications, so this entry makes none of those claims. Microsoft publishes separate documentation on data, privacy and security for Azure AI services that was not opened here. Custom endpoint models left unused for 7 days are decommissioned, which is a lifecycle fact from the page and not a data retention statement. Read the Azure data and privacy documentation before processing personal data.
Where Azure AI Speech falls short in practice
- Prices did not render. Per-hour rates for every tier were placeholders on the page we read, so no dollar figure could be verified and budgeting needs the live calculator.
- Free tier is narrow. F0 covers 5 audio hours a month of real-time transcription only, shared with Custom, and the Standard free tier does not support batch. A team that wants to test recorded files at no cost cannot do so on this tier.
- Add-ons multiply cost. Diarization, language identification and pronunciation assessment are each billed per hour, so a feature-rich pipeline costs a multiple of the base rate. Teams should price the complete pipeline, not only the base rate, before comparing with other vendors.
- Idle custom models disappear. Custom endpoint models unused for 7 days are decommissioned, so seasonal workloads must redeploy and re-test endpoints. That makes custom models awkward for seasonal or occasional workloads without a scheduled keep-alive job.
Works with
Alternatives to Azure AI Speech
Azure AI Speech: common questions
How much free transcription does Azure AI Speech include?
The free F0 tier includes 5 audio hours of real-time transcription per month, and that allowance is shared with Custom. The page states batch transcription is not supported on the Standard free tier, so recorded files cannot be processed within the free allowance.
What speech to text tiers does Azure list?
The pricing page lists real-time, Fast, Batch, MAI Transcribe-2 and -1.5, and LLM Speech Transcription under Standard, plus Custom real-time and batch under Custom. The rates for each tier were not rendered in our reading, so check the live page.
Does Azure charge extra for speaker labels?
Yes. The page lists diarization, continuous language identification and pronunciation assessment as enhanced add-ons priced per hour per feature. A pipeline using several add-ons pays for each one on top of the base transcription rate, so model the full stack before choosing.
What are the Azure commitment tiers?
The Custom section lists monthly commitment tiers for 2,000, 10,000 and 50,000 hours. The page text we captured gave no per-tier prices, so a buyer needs the live calculator or a Microsoft quote to see what the commitment saves. A calculator quote is safest.
What happens to unused Azure custom models?
The page says custom endpoint models unused for 7 days are decommissioned. A team with seasonal traffic should plan to redeploy and retest endpoints, or keep a low-volume job running. Endpoint hosting is also billed per model per hour, so idle endpoints are not free.