Tencent Cloud sells real-time speech in three separate pieces. Real-time ASR turns a stream into text. Real-time Speech Translation, built on its Hunyuan translation models, returns source and translated subtitles together. And a newer V2 ASR endpoint adds live speaker separation for Chinese and English. Each has its own engine list, price and concurrency cap, and the China site (cloud.tencent.com, billed in yuan) and the international site (tencentcloud.com, billed in dollars) price them very differently.
The LecSync Realtime Speech API takes the other route: one WebSocket returns the transcript, a translation between any two of 60 languages, speaker labels and live minutes, for $0.006 per audio minute ($0.36 per hour).
All Tencent figures below come from official pages read on October 18, 2026: the China-site ASR billing overview (updated October 16), the real-time ASR, real-time ASR V2 and real-time speech translation API docs, the Machine Translation billing page, and on the international site the ASR billing overview (updated October 15), real-time ASR and real-time speech translation docs and TMT billing. Yuan amounts are converted at 6.72 CNY per USD, the official central parity rate on October 16, 2026.
The short answer
If you are on the China site and need Mandarin or Chinese–English captions in the speaker's own language, Tencent is cheaper. Its "large model 2.0" engines cost ¥1.00 per hour pay-as-you-go (about $0.15), and a 1,000-hour pack is ¥950 (about $141). LecSync's $0.36 per hour does not beat that.
Bilingual captions are a different bill. Tencent's real-time speech translation costs ¥5.00 per hour in China (about $0.74) and $1.50 per hour on the international site, supports nine source languages, and starts at five concurrent streams per account. On tencentcloud.com, plain real-time ASR alone is $1.40 per hour. LecSync charges $0.36 per hour whether or not you turn on translation, speakers or minutes.
How each option 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 hands the returned stream URL and one-time token to the client, which sends 16 kHz mono PCM. Transcript segments, translations matched by segment ID, speaker labels with millisecond timestamps (speakers: true) and sections of minutes (digest: true) all come back on that connection. Direction is picked per segment, so people can switch between the two languages mid-meeting. After a network stall, backlog is capped at 2 seconds.
Tencent real-time ASR. The classic endpoint (wss://asr.cloud.tencent.com/asr/v2/<appid>) is chosen by engine_model_type: one engine per language or scenario, from 16k_zh and 16k_en to education, medical and court variants. The China docs list 17 languages plus dozens of Chinese dialects; the international docs list nine languages. The cheapest 2.0 engine on this endpoint, Hy-ASR-3.0-preview, is marked as a preview limited to 60-second inputs. Hot words (up to 128 per request), replacement lists and custom models are available.
Tencent real-time ASR V2. A separate endpoint whose 16k_zh_en_speaker_2.0 engine clusters voices on the fly and tags each sentence with a speaker_id from 0 to 9, starting at -1 until it is sure. It covers Chinese, English, Cantonese and about 30 dialects, and the docs place it on the 2.0 pricing scheme. It does not translate.
Tencent real-time speech translation. A third endpoint (/asr/speech_translate/) that returns source_text and target_text for each sentence, with an optional synthesized voice for Chinese or English targets (text-to-speech is billed separately). Source languages are Chinese, English, mixed Chinese–English, Japanese, Korean, Cantonese, Indonesian, Thai and Russian. The targets depend on the source: Russian, for example, can only go to Chinese or English. The result has no speaker field.
Side-by-side facts (checked October 18, 2026)
| LecSync Realtime Speech API | Tencent real-time ASR | Tencent real-time speech translation | |
|---|---|---|---|
| What one stream returns | Transcript, translation, speakers, minutes | Transcript (speakers on the V2 zh/en engine) | Source + translated subtitles; optional TTS |
| China-site list price | Same USD rate | ¥1.00/hr (2.0 engines), ¥3.20/hr (standard), ¥4.80/hr (1.0 large model) | ¥5.00/hr; 1,000 h pack ¥4,200 |
| International list price | $0.006/min ($0.36/hr), all features | $1.40/hr standard, $2.00/hr large model | $1.50/hr |
| Languages | 60, translated between any two | 17 + dialects (China), 9 (international) | 9 source codes, fixed target list per source |
| Live speaker labels | Every final segment | V2 16k_zh_en_speaker_2.0 only, up to 10 speakers | Not returned |
| Live minutes | digest: true on the same socket | No | No |
| Default concurrency | Not published | 200 (China classic), 50 (V2), 20 (international) | 5 |
| Billing unit | Accepted audio, fractions of a second | Per second, rounded up; tiers by hours per day | Same |
| Free tier | Pay as you go; a stream needs a $0.01 API balance | 5 h/month on standard engines (China site only) | None listed |
On the China site, Tencent's daily tiers are "reached" tiers: once a day's usage crosses a threshold, the whole day is billed at that tier's rate. Extra concurrency is sold monthly, for example ¥100 per channel for standard ASR and ¥280 per channel for speech translation.
What 1,000 hours cost
| 1,000 hours | LecSync | Tencent China site | Tencent international |
|---|---|---|---|
| Same-language captions | $360 | ¥950 pack (2.0 engine) ≈ $141; standard ¥1,800 ≈ $268 | $1,400 standard; $2,000 large model |
| Bilingual captions, built-in | $360 | Speech translation ¥4,200 pack ≈ $625 | $1,500 |
| Captions + translation + speaker labels | $360 | V2 speaker engine + Machine Translation: ≈ $607 (English speech) / ≈ $271 (Chinese speech), estimated | Not documented as one stream |
| Live minutes | $360 | Not offered | Not offered |
The DIY row assumes you translate only finished sentences with Tencent Machine Translation at ¥58 per million characters (each letter, space or Chinese character counts as one). English speech at around 150 words per minute is roughly 54,000 characters an hour; Chinese speech is closer to 15,000. Those volumes are estimates, so treat the row that way. The international TMT price is $10 per million characters. For the same math across other vendors, see what 1,000 hours cost once translation, speakers and summaries are added.
When Tencent is the better pick
- Mainland Chinese audio at volume. ¥1.00 per hour on the 2.0 engines is well under half of LecSync's rate, and Chinese-language speech also keeps a DIY translation pipeline cheap.
- Dialects and specialist engines. Engines for dozens of Chinese dialects, plus education, medical, court and phone-call variants, cover a lot of ground.
- High concurrency in China. The classic endpoint defaults to 200 streams per account, and extra channels cost ¥100 a month.
- Spoken output. Speech translation can read the translation aloud in Chinese or English. LecSync does not synthesize speech.
- Existing Tencent stack. If your audio, storage and billing already sit on cloud.tencent.com, one more product there is simple to approve.
When LecSync fits better
- Users outside mainland China. On tencentcloud.com, Tencent's cheapest real-time ASR is $1.40 per hour, almost four times LecSync, before translation.
- Language pairs beyond Tencent's list. Korean to Spanish, German to Arabic, Portuguese to Japanese: LecSync pairs any two of 60 languages. Tencent's translation endpoint takes nine source codes. Why this matters is covered in built-in translation vs a DIY STT + MT pipeline.
- Translation and speakers on one stream. At Tencent, live speaker IDs and live translation are separate endpoints; you would run both or stitch V2 output into Machine Translation. LecSync returns both on one connection. See building a meeting client with speakers and timestamps.
- Minutes during the call.
digest: truepushes confirmed sections while the meeting is still running. - One price to plan around. No engine tiers, no daily thresholds, no concurrency packs to buy. Stalls are capped at 2 seconds of backlog, as described in how LecSync caps backlog at 2 seconds.
FAQ
Is Tencent Cloud's China site priced the same as tencentcloud.com?
No. On October 18, 2026 the China site listed real-time ASR from ¥1.00 per hour on its 2.0 engines and ¥3.20 on standard engines, while the international site listed $1.40 per hour for standard real-time ASR and $1.50 for real-time speech translation. The China site also has a cross-border option for overseas end users at higher rates.
Does Tencent real-time ASR translate?
Not on the ASR endpoints. Translated subtitles come from the separate real-time speech translation endpoint, billed per hour of audio (¥5.00 in China, $1.50 internationally), or from your own pipeline that sends finished sentences to Tencent Machine Translation. LecSync includes translation on the transcription stream at the same $0.006 per minute.
Can Tencent label speakers in real time?
Yes, with limits. The V2 endpoint's 16k_zh_en_speaker_2.0 engine separates up to 10 speakers live for Chinese, English, Cantonese and Chinese dialects. Other engines and the speech translation endpoint don't return speaker IDs. LecSync labels every final segment in all 60 languages.
How much does an hour of LecSync cost?
$0.36. LecSync bills $0.006 per minute of accepted audio, counted to fractions of a second. That covers transcription in 60 languages, translation between any two, speaker labels, glossary, translation polish, live minutes and 30-day storage. Empty connections are not billed.
Bottom line
Pick Tencent Cloud for high-volume Mandarin or dialect captions on the China site, where its 2.0 engines run about $0.15 an hour, or when you need translated speech read aloud. Pick LecSync when your users are international, when language pairs go beyond Chinese, English and a handful of Asian languages, or when you want translation, speaker labels and running minutes on one stream at one price. A short test stream from the developers page with POST /v1/realtime/connect takes a few minutes to set up.