What Is the Cheapest Speech-to-Text API? Prices With Speaker Labels (2026)
On 2026-10-22, the cheapest pay-as-you-go speech-to-text API we found for pre-recorded audio with speaker labels was Soniox at $0.10 per hour, and the median of six vendors was $0.18. Deepgram was the highest at $0.26. MeetWave charges $0.15 with speaker labels and $0.20 with speaker names. In our test of 20.19 hours of recordings through the MeetWave API, files over 45 minutes came back in a median 214 seconds.
Key numbers
- Six speech-to-text APIs listed a median $0.18 per hour of pre-recorded audio with speaker labels on 2026-10-22, from $0.10 to $0.26 at pay-as-you-go list prices (How we calculated).
- On 2026-10-22, 3 of those 6 APIs stated that speaker labels cost nothing extra on pre-recorded audio, one charged $0.02 per hour for them, and two listed no separate batch charge (How we calculated).
- Free allowances at sign-up ranged from none stated to $200 of credit across the same 6 APIs on 2026-10-22; two of them give monthly minutes instead of credit (How we calculated).
- 20.19 hours of meetings and earnings calls (26 files) sent through the MeetWave API on 2026-10-22 came back in a median 214 seconds for files over 45 minutes and 130 seconds for files of 15 to 45 minutes, and cost $3.03 at $0.15 per hour (How we calculated).
- The same 2026-10-22 test scored 23.7% word error rate on 16 AMI meeting recordings and 9.6% on 10 Earnings-22 earnings calls (How we calculated).
The price list with speaker labels
Soniox is the cheapest list price for pre-recorded audio with speaker labels, at $0.10 per hour with diarization bundled in (Soniox pricing). AssemblyAI's Universal-2 is $0.15 per hour plus $0.02 for speaker diarization, $0.17 in total (AssemblyAI pricing). Deepgram's Nova-3 is $0.0043 per minute, $0.26 per hour, with diarization included on pre-recorded audio (Deepgram pricing). MeetWave charges $0.15 per hour with speaker labels and $0.20 with speaker names (MeetWave API pricing). Figure 1 puts all seven side by side.
Figure 1. List price per hour of pre-recorded audio with speaker labels, pay as you go. Source: the vendors' own pricing pages, read 2026-10-22.
| API | Pre-recorded, per hour | Speaker labels | Free at sign-up | Billing |
|---|---|---|---|---|
| Soniox (async) | $0.10 | Included | Not stated | Not stated |
| MeetWave | $0.15 ($0.12 without labels, $0.20 with names) | Included in the $0.15 tier | 1 hour, 60 days | Per second, no minimum |
| AssemblyAI (Universal-2) | $0.15 | +$0.02 | $50 credit | Not stated |
| Azure AI Speech (batch) | $0.18 | No batch charge listed | 5 hours a month, real-time only | Not stated |
| Google Cloud (V2 dynamic batch) | $0.18 | No charge listed | 60 minutes a month (V1 API) | Per second |
| Speechmatics (Melia 1) | $0.24 | Included | $100 credit | Per second |
| Deepgram (Nova-3) | $0.26 | Included | $200 credit | Not stated |
Source: Soniox, MeetWave, AssemblyAI, Azure, Google Cloud, Speechmatics, Deepgram pricing pages, read 2026-10-22.
The big cloud prices depend on the mode. Google Cloud's V2 standard recognition is $0.016 per minute for the first 500,000 minutes a month, $0.96 per hour, while dynamic batch, which processes audio at a lower urgency, is $0.003 per minute, $0.18 per hour (Google Cloud pricing). Azure lists batch transcription at $0.18 per hour and real-time at $1 per hour, with enhanced add-on features, diarization among them, priced for real-time (Azure retail prices, Azure pricing).
What "free" means on each API
Free credit is the largest at Deepgram, $200 for new accounts, then Speechmatics at $100 with no card required and AssemblyAI at $50 (Deepgram pricing, Speechmatics pricing, AssemblyAI pricing). Google Cloud gives 60 free minutes a month on its V1 API, and Azure 5 free audio hours a month for real-time transcription only; batch is not supported on Azure's free tier (Google Cloud pricing, Azure pricing). MeetWave gives one free hour, valid for 60 days (MeetWave API pricing).
At these prices, $200 of credit is about 775 hours of Deepgram Nova-3 pre-recorded audio, and one free hour is one meeting (How we calculated). For a test of your own pipeline, the credit size decides how much you can try before paying.
What a list price leaves out
The per-hour rate is only one part of the bill. Speechmatics and Google Cloud bill to the second, and Google rounds each request up to the next second (Speechmatics pricing, Google Cloud pricing). Speechmatics takes 33% off if you let it train on your audio, and keeps that off by default (Speechmatics pricing). Live audio costs more: Soniox streams at $0.12 per hour against $0.10 for files, and Azure charges $1 per hour for real-time against $0.18 for batch (Soniox pricing, Azure retail prices).
What the API returned in our test
We sent the 26 recordings of a public benchmark through the MeetWave API on 2026-10-22 the way a customer would: 16 AMI meetings uploaded as files and 10 Earnings-22 earnings calls passed as links, with speaker labels on (How we calculated). The 20.19 hours cost $3.03 at the $0.15 tier. Files of 15 to 45 minutes came back in a median 130 seconds, 154 at the 90th percentile, and files over 45 minutes in a median 214 seconds, 245 at the 90th percentile (How we calculated). Figure 2 shows the three length groups.
Figure 2. Median time from submitting a file to a finished transcript with speaker labels, by file length. Source: MeetWave measurement through the public API, 26 recordings, 2026-10-22.
Speed and price say nothing about the words, so the same run was scored for accuracy. It reached 23.7% word error rate on the AMI meetings and 9.6% on the earnings calls, within 0.1 percentage point of the same files run on 2026-09-23 (How we calculated). Meetings are hard audio for every engine; how engines compare on them is in our transcription accuracy research.
How we calculated
Which APIs. The comparison covers six vendors: AssemblyAI, Deepgram, Google Cloud and Azure, four of the five APIs in our transcription pricing research, plus Speechmatics and Soniox. MeetWave is shown beside them and left out of the median.
Prices. List prices for pre-recorded (batch) audio on each vendor's public pricing page, read on 2026-10-22, pay as you go, without volume discounts, commitments or savings plans. A per-minute price is multiplied by 60 and rounded to the cent: Deepgram $0.0043 × 60 = $0.258, shown as $0.26; Google Cloud dynamic batch $0.003 × 60 = $0.18 (Deepgram pricing, Google Cloud pricing). "With speaker labels" adds a diarization add-on where one is sold separately: AssemblyAI $0.15 + $0.02 = $0.17 (AssemblyAI pricing). Google Cloud and Azure list no batch diarization charge, so their base batch price is shown. Azure's batch price comes from Microsoft's Retail Prices API, meter "S1 Speech to Text Batch" in East US, because the pricing page shows amounts only after a region is chosen (Azure retail prices).
Median. Over the six vendors, MeetWave excluded: $0.10, $0.17, $0.18, $0.18, $0.24 and $0.26, so the median is ($0.18 + $0.18) ÷ 2 = $0.18. Deepgram's $200 of credit buys $200 ÷ $0.258 ≈ 775 hours of Nova-3 pre-recorded audio (Deepgram pricing).
Measurement. On 2026-10-22 a script sent 26 files through the MeetWave public API on its staging copy, which runs the same code and prices: the 16 meetings of the AMI Meeting Corpus test split, Mix-Headset audio, 9.06 hours, CC BY 4.0, uploaded as files; and the 10 earnings calls of the Earnings-22 subset10, 11.12 hours, CC BY-SA 4.0, passed as links (Del Rio et al., 2022). Each request had speaker labels on and the language detected. Turnaround is the job's own created_at to completed_at. Word error rate is (substitutions + deletions + insertions) ÷ reference words, summed over each set, after both texts pass through the Open ASR Leaderboard's English text normaliser, scored with the open-source jiwer library. The files, references and normaliser are those of our 2026-09-23 run, which scored 23.7% on AMI and 9.5% on Earnings-22.
FAQ
Which speech-to-text API is the cheapest?
For pre-recorded audio with speaker labels, Soniox at $0.10 per hour was the lowest list price we found on 2026-10-22; the median of six vendors was $0.18 (Soniox pricing, How we calculated).
Is there a free speech-to-text API?
Not without limits. Deepgram gives $200 of credit, Speechmatics $100 and AssemblyAI $50, and Google Cloud 60 minutes a month on its V1 API (Deepgram pricing, Speechmatics pricing, AssemblyAI pricing, Google Cloud pricing).
Do speaker labels cost extra?
At AssemblyAI they add $0.02 per hour; Deepgram, Speechmatics and Soniox include them in the pre-recorded price (AssemblyAI pricing, Deepgram pricing, Speechmatics pricing, Soniox pricing).
How fast is a cheap API?
In our 2026-10-22 test, files of 15 to 45 minutes came back in a median 130 seconds and files over 45 minutes in a median 214 seconds, with speaker labels on (How we calculated).
Sources
- Soniox — Pricing (2026). Accessed 2026-10-22.
- AssemblyAI — Pricing (2026-09-28). Accessed 2026-10-22.
- Deepgram — Pricing (2026). Accessed 2026-10-22.
- Speechmatics — Pricing (2026). Accessed 2026-10-22.
- Google Cloud — Speech-to-Text pricing (2026). Accessed 2026-10-22.
- Microsoft — Azure AI Speech pricing (2026). Accessed 2026-10-22.
- Microsoft — Azure Retail Prices API, Azure Speech, East US (2026). Accessed 2026-10-22.
- MeetWave — Speech-to-text API pricing (2026). Accessed 2026-10-22.
- Del Rio et al. — Earnings-22: A Practical Benchmark for Accents in the Wild (2022). Accessed 2026-09-23.
More on this: API docs
