Amazon Transcribe streaming is a reasonable default for live captions if your backend already runs on AWS. The gap appears when the product also needs a second language on screen. Transcribe returns text in the language that was spoken; translation is a different service, Amazon Translate, with its own API and its own per-character bill. The LecSync Realtime Speech API returns captions, translation and speaker labels on one WebSocket for $0.006 per audio minute.
This comparison uses public sources only. AWS prices come from the AWS Price List API for US East (N. Virginia) and the Amazon Transcribe pricing page, read on October 10, 2026. LecSync figures come from the developers page and the pricing docs.
The short answer
If you need live captions, a translation track and speaker labels together, LecSync returns all three over one connection at $0.36 per hour (60 × $0.006). Amazon Transcribe standard streaming lists at $0.0001667 per second in US East, about $0.60 per hour, and speaker diarization is included in that price. Translation is extra: Amazon Translate charges $15 per million characters, which works out to roughly $0.75 for an hour of English speech. Together that is about $1.35 per hour.
Amazon Transcribe is the better fit when your security and billing controls already live in AWS, or when you need PII redaction, custom language models, Call Analytics, or audio formats beyond 16 kHz mono PCM.
What you wire up on each side
With LecSync, your server makes one POST to /v1/realtime/connect and gets back a stream URL plus a one-time token. The client opens the WebSocket and pushes 16 kHz mono PCM. Transcript segments, their translations, speaker labels and optional live minutes all come back on that same socket, already paired.
With AWS, you call StartStreamTranscription over HTTP/2 or a SigV4-signed WebSocket URL. Transcript events arrive in the spoken language, with speaker labels if you set ShowSpeakerLabel. To show a translation, your service takes each final segment, sends it to Translate's TranslateText (up to 10,000 bytes per request), and merges the result back onto the caption timeline. That is two services, two meters, and the glue code is yours to maintain. We covered that pattern in more depth in built-in translation vs a DIY STT + MT pipeline.
Side-by-side facts (checked October 10, 2026)
| LecSync Realtime Speech API | Amazon Transcribe streaming | |
|---|---|---|
| Connection | One WebSocket after POST /v1/realtime/connect | HTTP/2 or WebSocket via StartStreamTranscription |
| Live captions | Included | $0.0001667/sec ≈ $0.60/hr (US East list) |
| Live translation | Included; one language pair per stream, direction chosen per segment | Not in Transcribe; Amazon Translate at $15 per million characters |
| Speaker labels | speakers: true on the same stream | ShowSpeakerLabel, included in the standard price; up to 30 speakers |
| Live minutes / summary | digest: true, pushed during the session | Not offered for standard streaming; call summarization is post-call in Call Analytics |
| Mixed-language audio | Third language that appears is translated into your target | IdentifyMultipleLanguages with at least two language codes; transcript only |
| Custom terms | Glossary terms, forced translation pairs and context, included | Custom vocabularies included; custom language models are a paid add-on |
| Audio input | pcm_s16le, 16 kHz, mono (v1) | PCM, FLAC, Ogg Opus; two-channel audio supported |
| Default concurrency | Not published on the developers page | 25 concurrent streams per Region, adjustable |
| Long sessions | Connect token only limits time to open the socket; server-side backlog capped at 2 seconds | Hard maximum session duration; SessionResumeWindow allows reconnects for up to 300 minutes from stream start |
| Billing grain | Accepted audio, fractions of a second | One-second increments, no minimum for standard streaming |
| Free allotment | None on the public rate | 60 minutes/month for 12 months; Translate adds 2M characters/month for 12 months |
| SLA | Best-effort, no SLA today | Check AWS's service terms for Transcribe |
Sources: Amazon Transcribe pricing, Amazon Translate pricing, speaker diarization, endpoints and quotas, streaming language identification, StartStreamTranscription reference, LecSync connect docs.
What an hour, and 1,000 hours, actually cost
Translate bills by characters, so the AWS total depends on how much text an hour of speech produces. We assume about 50,000 characters per hour, roughly 150 words a minute of English. That assumption is ours; your transcripts may run longer or shorter.
| LecSync | Amazon Transcribe + Translate | |
|---|---|---|
| Captions | Included | $0.60/hr |
| Speaker labels | Included | Included |
| Translation | Included | ~$0.75/hr (50,000 chars × $15/M) |
| Live minutes | Included | Not available in streaming |
| Per hour | $0.36 | ~$1.35 |
| 1,000 hours | $360 | ~$1,350 ($600 + $750) |
In the first year, the AWS free tiers take a little off: 60 Transcribe minutes and 2 million Translate characters a month bring the 1,000-hour figure to about $1,319. Character counts also vary by language. Chinese or Japanese source text produces far fewer characters per hour than English, so the Translate line shrinks for those speakers.
Captions alone tell a similar story. Transcribe streaming at about $0.60 per hour is still above LecSync's $0.36 with everything switched on. Transcribe's batch rate of $0.006 per minute matches LecSync's price, but batch works on uploaded files, so it does not help a live caption screen. For a wider look at add-on pricing across vendors, see what 1,000 hours cost with translation, speakers and summaries.
Where Amazon Transcribe is the better pick
- Your platform is AWS. IAM, consolidated billing and Service Quotas already work. The endpoint list includes FIPS endpoints and GovCloud Regions.
- You handle sensitive data. Streaming can identify and redact PII, priced as an add-on ($0.0024 per minute at the first US East tier).
- You need a trained model. Custom language models are available for streaming, alongside custom vocabularies and vocabulary filters.
- You run a contact center. Call Analytics adds real-time sentiment, issue detection and category events, with post-call summaries. Transcribe Medical covers US English clinical speech.
- Your audio isn't 16 kHz mono PCM. Transcribe accepts FLAC and Ogg Opus and can transcribe two channels separately.
- You're testing at low volume. The free tier covers 60 minutes a month for a year.
Where LecSync fits better
- Captions and translation on the same screen. Translation streams alongside the transcript and adds about 88 ms to the first token, with no second service to call. Any of 60 languages can be translated into any other.
- Speakers and minutes without extra jobs. Speaker labels come with timestamps, and live minutes build up section by section while the session runs. See live minutes vs post-call summarization.
- Thin clients. A one-time token opens the socket, so a browser or mobile app never signs AWS requests.
- Latency that doesn't drift. First words reach the screen in about 0.28 seconds, and the server caps backlog at 2 seconds after a network stall.
- One line on the invoice. $0.006 per accepted audio minute, whether you enable one feature or all of them.
The trade-offs are real. LecSync takes one language pair per stream, accepts only 16 kHz mono PCM in v1, and offers no SLA today.
FAQ
Does Amazon Transcribe streaming translate speech?
No. Transcribe returns text in the language spoken. With multi-language identification it can produce a transcript that mixes languages, and AWS's own docs point to other services for translation. The usual pairing is Amazon Translate, billed at $15 per million characters for standard text translation.
How much does Amazon Transcribe streaming cost per hour?
The AWS Price List shows $0.0001667 per second for standard streaming in US East (N. Virginia), about $0.60 per hour, billed per second with no minimum. Custom vocabularies, vocabulary filtering, speaker diarization and language identification are included. PII redaction and custom language models cost extra.
Does Amazon Transcribe streaming support speaker labels?
Yes. Set ShowSpeakerLabel on StartStreamTranscription. Amazon Transcribe can tell apart up to 30 speakers, labelled spk_0 through spk_29. Diarization is part of the standard streaming price.
How many streams can I run at once on Amazon Transcribe?
The default quota is 25 concurrent streams per Region across HTTP/2 and WebSocket, and you can request an increase in the Service Quotas console. AWS also warns that ramping up streams too fast during load tests can trigger throttling.
What does LecSync include in $0.006 per minute?
Realtime transcription in 60 languages, realtime translation, audio-aligned timestamps, speaker separation, glossary and domain context, AI translation refinement, live structured minutes, and optional 30-day storage of recordings and transcripts. The rate stays the same whichever of these you turn on.
What happens when a speaker switches languages mid-lecture?
On AWS, IdentifyMultipleLanguages needs a list of at least two expected languages and returns a transcript in each language detected; translating it is a separate step. On LecSync, the stream is set to one language pair, direction is decided per segment, and a third language that turns up is translated into your target language.
The short version
Choose by whether you want to run the translation layer yourself. For captions, translation and speakers on one socket at $0.36 per hour, LecSync is the simpler build. If you need AWS-native controls, PII redaction, custom models or call analytics, Amazon Transcribe plus Translate is the stronger fit at roughly $1.35 per hour. Create a key in the LecSync developers console and follow the connect docs to open your first stream. If you're also weighing other clouds, compare Google Cloud Speech-to-Text and Azure AI Speech.