IBM Watson Speech to Text streams over a WebSocket, returns interim and final results, and can label up to six speakers. It is a mature service with custom language and acoustic models, telephony models for 8 kHz call audio, and the option to run it on your own infrastructure. It does not translate speech, and IBM withdrew its Watson Language Translator service in December 2024.
The LecSync Realtime Speech API covers that gap on a single connection: preview and final captions, translation between any two of 60 languages, a speaker and audio-aligned timestamps on every final segment, and running minutes. It costs $0.006 per audio minute ($0.36 per hour) with everything switched on.
The IBM figures below were checked on October 22, 2026, against the IBM Cloud catalog price list for the Plus plan, the pricing FAQ, and IBM's docs on the WebSocket interface, models, large speech models and speaker labels.
The short answer
If you run a call center, need 8 kHz phone models, or must deploy speech recognition on premises, IBM Watson is a reasonable pick. Its Plus plan lists $2.12 per 100 minutes in the IBM Cloud catalog for October 2026, which is $0.0212 per minute, or $1,272 for 1,000 hours. IBM's pricing FAQ still shows the older round figure of $0.02.
For translated captions, lectures or meetings with several speakers, or anything you want summarized while it happens, LecSync does more for less. 1,000 hours cost $360, about 72% below IBM's Plus list price, and translation is already in that number.
How each API works
LecSync. Your server calls POST https://api.lecsync.com/v1/realtime/connect with the audio format, a language pair and the features you want, then streams 16 kHz mono PCM to the URL it gets back. Preview text arrives while someone is still talking, then a final segment once the sentence settles. Translations come paired with the segment ID. With speakers: true, each final carries a speaker label and startMs/endMs on the recording's clock. With digest: true, minutes arrive section by section. The first word shows up about 0.28 seconds after speech starts, and translation adds roughly 88 ms.
IBM Watson Speech to Text. You open wss://api.{location}.speech-to-text.watson.cloud.ibm.com/instances/{instance_id}/v1/recognize with a token and a model, send a JSON start message, then send audio as binary frames. Set interim_results: true to get partial text before finals. low_latency: true speeds results up on most next-generation models, and on the en-US large speech model since June 2026. IBM notes that leaving it off usually gives more accurate text.
Two connection details catch people building caption apps:
- Silence can close the session. By default the service ends a session after 30 seconds of audio with no detected speech. You can set
inactivity_timeoutto-1, and the docs say silence you send to keep a session alive is billed. - You pick the model, and the language with it. Models are named by locale and audio type, such as
en-US_Multimediaorzh-CN_Telephony. IBM has a separate/v1/detect_languageendpoint, but a recognition stream does not switch languages on its own.
Languages and models
IBM now sells three families: large speech models, next-generation telephony models for 8 kHz audio, and next-generation multimedia models for 16 kHz audio. Previous-generation models were discontinued in August 2023.
Large speech models cover English, French, German, Spanish, Portuguese, Japanese, Dutch and Italian, with Dutch added in April 2026 and Italian in May 2026. The next-generation telephony list adds Arabic, Mandarin Chinese, Czech, Hindi, Korean and Swedish, for 14 languages in total. Mandarin is offered only as a telephony model, with no multimedia or large speech model.
LecSync recognizes 60 languages, detects which one is being spoken, and translates between any two of them on the same stream.
Speaker labels
Set speaker_labels: true and IBM attaches a speaker number to each word, using word timestamps. The docs are open about where it works best: two people on a phone call. It handles up to six speakers, and accuracy can be poor with one dominant voice, as in a lecture or podcast, or when people talk over each other. The first 30 seconds are weaker, and results usually settle after about a minute of audio. Speaker IDs can change until the final result, and with interim results on, final labels for long streams can arrive well after the text.
LecSync attaches a speaker and millisecond timestamps to every final segment, and those times stay on one recording clock across reconnects. See building a meeting client with speakers and timestamps.
Translation and minutes
IBM Speech to Text returns transcripts only. Watson Language Translator, which used to fill the translation step, was withdrawn on December 10, 2024. For live translated captions you would send each final sentence to another translation service or a language model, then match each result to its line yourself. The work involved is covered in built-in translation vs a DIY STT + MT pipeline.
IBM also offers enrichments=punctuation to clean up punctuation and capitalization. It has no live summary or minutes feature. With LecSync, digest: true sends confirmed sections during the session; see live digest vs post-call summarization.
Side-by-side facts (checked October 22, 2026)
| LecSync Realtime Speech API | IBM Watson Speech to Text (Plus) | |
|---|---|---|
| Price | $0.006/min ($0.36/hr), all features | $2.12 per 100 min ($0.0212/min) in the catalog; $0.0106/min from 1M min a month |
| Free tier | Not listed on the developers page | Lite: 500 min a month |
| Languages | 60, auto-detected | 14 across model families; one locale per stream |
| Mandarin Chinese | Yes, with translation | Telephony model only |
| Live translation | Any two of 60, same socket | None; Language Translator withdrawn Dec 2024 |
| Interim results | Preview, then final | interim_results, optional low_latency |
| Speaker labels | Every final segment, with startMs/endMs | Per word, up to 6, tuned for two-person calls |
| Live minutes | digest: true | No |
| Custom vocabulary | Glossary and domain context | Custom language and acoustic models, grammars, no extra charge |
| Concurrency | Not published | Up to 100 streams on Plus |
| Regions | 7 gateway regions | 7 locations in the docs, including Seoul |
| Self-hosting | No | Yes, through IBM Cloud Pak for Data and Software Hub |
| Uptime commitment | Best effort, no SLA | Covered by IBM Cloud service level agreements |
IBM's catalog lists six locations for the Plus plan, without Seoul. Premium adds data isolation, bring-your-own-key encryption, HIPAA readiness and 150,000 free minutes a month, and it is sold through IBM sales.
What 1,000 hours cost
| Workload, 1,000 hours a month | LecSync | IBM Watson Plus |
|---|---|---|
| Same-language captions | $360 | $1,272 |
| Translated captions | $360 | $1,272 plus a separate translation service |
| Speaker labels with timestamps | Included | Included, per word |
| Live minutes | Included | Not available |
1,000 hours is 60,000 minutes, multiplied by $0.006 or $0.0212. At IBM's own $0.02 FAQ rate the total would be $1,200. Even past one million minutes a month, IBM's $0.0106 works out to about $0.64 an hour, still above LecSync's $0.36. Other vendors priced the same way are in what 1,000 hours cost once translation, speakers and summaries are added.
When IBM Watson is the better pick
- Phone audio. Telephony models are built for 8 kHz call audio, and speaker labels are tuned for two-person calls.
- Deep customization. Custom language and acoustic models and grammars at no extra charge on Plus.
- On-premises or private cloud. The same service runs on IBM Cloud Pak for Data and IBM Software Hub.
- Compliance and contracts. Premium offers data isolation, BYOK and HIPAA readiness, and IBM Cloud publishes SLAs with service credits. LecSync has no SLA today.
- Already on IBM. If watsonx Assistant or other IBM services already run your stack, staying on one bill may matter more.
When LecSync fits better
- Translated live captions. Any pair of 60 languages, on the same socket, each translation tied to its segment.
- Chinese and mixed-language rooms. Mandarin with translation, and the language is detected per segment instead of fixed per stream.
- Lectures and meetings. Speaker labels and timestamps on every final, without a two-speaker assumption.
- Long, unstable sessions. Server-side backlog is capped at 2 seconds, so captions catch up within a second after a slowdown, as described in how LecSync caps backlog at 2 seconds.
- Price. $0.36 an hour with translation, speakers and minutes, against $1.27 an hour for IBM transcription alone.
Comparing big cloud vendors? See LecSync vs Google Cloud Speech streaming, LecSync vs Azure Speech realtime translation and LecSync vs AWS Transcribe streaming.
FAQ
How much does IBM Watson Speech to Text cost?
As of October 22, 2026, the IBM Cloud catalog lists the Plus plan at $2.12 per 100 minutes, or $0.0212 per minute, dropping to $0.0106 after one million minutes in a month. IBM's pricing FAQ still shows $0.02 and $0.01. The Lite plan includes 500 free minutes a month. Premium pricing comes from IBM sales.
Does IBM Watson Speech to Text translate speech in real time?
No. The service returns transcripts only, and IBM withdrew Watson Language Translator on December 10, 2024. For live translated captions you need a separate translation service or language model and your own code to pair each translation with its line. LecSync returns translations on the same connection.
How many speakers can IBM Watson identify?
Up to six, with speaker_labels: true. IBM says the feature is optimized for two-person telephone conversations and can be inaccurate with one dominant speaker, cross-talk or short turns. Speaker IDs can change until final results. LecSync labels every final segment with a speaker and timestamps.
Does IBM Watson support Chinese streaming recognition?
Yes, through the zh-CN_Telephony next-generation model, which supports low latency. IBM lists no multimedia or large speech model for Mandarin, so it is aimed at phone audio. LecSync handles Mandarin at 16 kHz and can translate it to and from any of its 60 languages.
Bottom line
Choose IBM Watson for call-center audio, heavy model customization, on-premises deployment or a contract with an SLA. Choose LecSync for translated captions, lectures and meetings with several speakers, Chinese and mixed-language audio, and minutes during the session, at $0.006 per minute. You can open a test stream from the developers page with one POST /v1/realtime/connect.