# Paxa Labs and Amazon Transcribe for multilingual speech to text

A transcription API for Thai and other languages next to Amazon Transcribe, across Thai mixed with English, language support, how the transcript comes back, speakers, vocabulary, subtitles, and live transcription.

Amazon Transcribe lists Thai and Simplified Chinese for batch and streaming. Its multi-language identification transcribes each language when a speaker switches mid-conversation, and its batch jobs read each recording from an Amazon S3 bucket. Paxa STT Lite transcribes Thai, English, Mandarin, and other languages, and one request returns the transcript in its response. A Thai sentence that turns into English is written as it was spoken, with no setting. Paxa STT Lite is a Research preview. Requests bill at the published rate, and its transcript quality and options may change between releases.

Sources: [Amazon Transcribe supported languages](https://docs.aws.amazon.com/transcribe/latest/dg/supported-languages.html), [Amazon Transcribe batch language identification](https://docs.aws.amazon.com/transcribe/latest/dg/lang-id-batch.html)

Rates and claims last checked October 2026

## What differs

| Dimension | Paxa Labs | [Amazon Transcribe](https://aws.amazon.com/transcribe/pricing/) |
| --- | --- | --- |
| Thai with English inside it | A Thai sentence that switches into English is written as spoken, with no setting. Nothing is translated. | Multi-language identification transcribes each language a batch file contains, with th-TH, zh-CN, and en-US on its list. Redaction and custom language models do not work with it. |
| Languages | Thai, English, Mandarin, and other languages. Paxa STT Lite writes speech in the language spoken and detects that language itself. A BCP 47 hint corrects a recording it misreads. | Thai (th-TH) and Simplified Chinese (zh-CN) for batch and streaming. Their table lists no custom language model or redaction for either, and no Call Analytics for Thai. |
| Getting the transcript | One request with model paxa-stt-lite-v1-preview returns the transcript in its response, for recordings up to 120 minutes and 95 MiB. | A batch job reads the media file from an Amazon S3 bucket, up to 28,800 seconds and 2 GB per file. |
| Speaker turns | Speaker turns numbered from 0 in order of first appearance, on recordings up to 120 minutes, at no extra charge. | Up to 30 speakers, labeled spk_0 through spk_29. |
| Vocabulary | A vocabulary of up to 50 terms, in Thai or any other language, at no extra charge. | A custom vocabulary file of up to 50 Kb, created ahead of time in the same AWS Region as the job. It works with every supported language. |
| Subtitles | An SRT or WebVTT file in the same response. Every line break falls on a word boundary, Thai included. | WebVTT and SubRip files from a batch job, stored beside the transcript in Amazon S3. Numbering starts at 0 unless the request sets 1. |
| Live transcription | Paxa STT Lite Realtime streams over a WebSocket, reports when the user starts and stops talking, and sends one final transcript per turn. | Streaming over HTTP/2 or a WebSocket, in FLAC, Ogg Opus, or PCM, with 25 concurrent streams per Region by default. |

Sources: [Amazon Transcribe batch language identification](https://docs.aws.amazon.com/transcribe/latest/dg/lang-id-batch.html), [Amazon Transcribe supported languages](https://docs.aws.amazon.com/transcribe/latest/dg/supported-languages.html), [Amazon Transcribe data input and output](https://docs.aws.amazon.com/transcribe/latest/dg/how-input.html), [Amazon Transcribe quotas](https://docs.aws.amazon.com/general/latest/gr/transcribe.html), [Amazon Transcribe speaker diarization](https://docs.aws.amazon.com/transcribe/latest/dg/diarization.html), [Amazon Transcribe custom vocabularies](https://docs.aws.amazon.com/transcribe/latest/dg/custom-vocabulary.html), [Amazon Transcribe subtitles](https://docs.aws.amazon.com/transcribe/latest/dg/subtitles.html), [How Amazon Transcribe works](https://docs.aws.amazon.com/transcribe/latest/dg/limits-guidelines.html)

- Paxa STT Lite, per hour of audio: $0.50
- Amazon Transcribe: $0.36
- Their rate, as a multiple: 0.72×

Amazon Transcribe's rate is standard batch transcription in US East (N. Virginia), billed by the second with no minimum, before redaction, custom language models, and Call Analytics. Every rate is per hour of recorded audio, from the vendor's cheapest pay-as-you-go model to its most expensive. Per-minute and per-second rates are converted to that unit.

## What it costs at your volume

| Audio, per month | Paxa Labs | Amazon Transcribe | Difference |
| --- | --- | --- | --- |
| 10 hours | $5 | $3.60 | $1.40 more |
| 100 hours | $50 | $36 | $14 more |
| 1,000 hours | $500 | $360 | $140 more |
| 10,000 hours | $5,000 | $3,600 | $1,400 more |

Every figure is the two published rates multiplied by the volume beside it. Where the other vendor publishes a range, the low end is compared and the range is shown.

## When Amazon Transcribe is the better choice

Ours is a Research preview, and its output and options may change between releases. A team that needs a settled model today has a reason to stay. Amazon Transcribe takes a file up to 2 GB and 28,800 seconds, and ours stops at 120 minutes. For English it adds redaction, custom language models, and Call Analytics, and its batch jobs read audio already stored in Amazon S3. Choose us when Thai speakers switch into English and one request should return the transcript.

## Questions

### Does Amazon Transcribe support Thai?

Yes. Their supported languages table lists Thai (th-TH) for batch and streaming, and multi-language identification includes it. The same table lists no custom language models, redaction, or Call Analytics for Thai.

### What does Research preview mean for Paxa STT Lite?

Requests bill at the published rate. Transcript quality and the option set may change between releases, and the changelog on our speech to text page records each change.

### Do you publish accuracy or latency comparisons?

No. We publish no accuracy, quality, or latency figures of our own, for our model or anyone else's. Transcribe the same recording on both and read the two transcripts.

### How current are the prices here?

They come from each vendor's published list rate, recorded in one place and re-checked on a fixed schedule. The month of the last check is at the top of the page.

## Other comparisons

- [Best Speech to Text APIs for Each Job](https://paxalabs.com/compare/speech-to-text-apis.md)
- [Paxa Labs and Deepgram for multilingual speech to text](https://paxalabs.com/compare/deepgram.md): $0.26-0.31
- [Paxa Labs and AssemblyAI for multilingual speech to text](https://paxalabs.com/compare/assemblyai.md): $0.15-0.21
- [Paxa Labs and OpenAI for multilingual speech to text](https://paxalabs.com/compare/openai-speech-to-text.md): $0.27-0.36
- [Paxa Labs and Google Cloud Speech-to-Text for multilingual transcription](https://paxalabs.com/compare/google-speech-to-text.md): $0.18-0.96
- [Paxa Labs and Speechmatics for multilingual speech to text](https://paxalabs.com/compare/speechmatics.md): $0.24-0.75
- [Paxa Labs and ElevenLabs Scribe for multilingual transcription](https://paxalabs.com/compare/elevenlabs-scribe.md): $0.22
- [Paxa Labs and Soniox for Thai speech to text](https://paxalabs.com/compare/soniox.md): $0.10
- [Paxa Labs and Gladia for multilingual speech to text](https://paxalabs.com/compare/gladia.md): $0.18-0.30
- [Paxa Labs and Rev AI for multilingual speech to text](https://paxalabs.com/compare/rev-ai.md): $0.20-0.30

## Start with free credits

Sign in with Google, GitHub, or Hugging Face and spend 100 one-time free credits on your own text.