LecSync vs Volcengine Doubao Realtime ASR: Captions, Interpretation and Pricing

ByteDance sells its Doubao speech models in two places: Volcengine for mainland China, billed in yuan, and BytePlus for everyone else, billed in dollars out of Singapore. Both run the same family of products — Seed ASR 2.0 for streaming transcription and a separate Live Interpretation model that turns speech into translated subtitles or dubbed audio. The LecSync Realtime Speech API costs $0.006 per audio minute ($0.36 per hour) and returns the transcript, a translation between any two of 60 languages, speaker labels and live minutes over one WebSocket.

On raw transcription, ByteDance is cheaper, and this article says so plainly. The decision gets more interesting once you need a second language, labelled speakers, or notes while the call is still going.

All figures come from public pages read on October 17, 2026: the Volcengine Doubao Speech billing page (updated September 29), its streaming ASR API reference and Simultaneous Interpretation 2.0 API; on BytePlus, the ASR billing page, Live Interpretation billing, the Live Interpretation API guide and the ASR 2.0 feature table. Yuan figures are converted at an estimated 7.0 CNY per USD.

The short answer

For captions in the speaker's own language, Doubao streaming ASR 2.0 is one of the cheapest list prices around: ¥1 per hour pay-as-you-go in China (about $0.14), or $0.15 per hour on BytePlus. A 1,000-hour prepaid pack is ¥900 or $140. LecSync's $0.36 per hour does not compete on that line.

For bilingual captions, ByteDance points you to a different product. Live Interpretation is billed by tokens. Input audio alone works out to ¥1.80 or $0.27 per hour, and translated text is metered on top. One side of every language pair has to be Chinese or English. LecSync keeps translation on the same stream and the same $0.36 rate, for any pair among 60 languages, with speaker labels and minutes included.

How each option works

LecSync. Your backend calls POST https://api.lecsync.com/v1/realtime/connect with the audio format, one language pair and the features you want, and receives a stream URL plus a one-time token. The client streams 16 kHz mono PCM. Source segments, translations paired by segment ID, speaker labels with millisecond timestamps (speakers: true) and sections of minutes (digest: true) all come back on that connection. Translation direction is decided per segment, so a Chinese–English meeting where people switch languages still lands in the right column. Network stalls are capped at 2 seconds of backlog.

Doubao streaming ASR 2.0. There are three WebSocket endpoints. The bidirectional endpoints return text word by word; turning on enable_nonstream adds a second pass that re-recognizes each utterance after 800 ms of silence and marks it final. The streaming-input endpoint (bigmodel_nostream) is the only one that accepts an explicit language code (about 25 languages on the China docs; BytePlus advertises 51 for ASR 2.0), and it waits for 15 seconds of audio or the last packet before returning text. Hot words and context work on the Chinese–English model.

Doubao Live Interpretation 2.0. A separate socket (/api/v4/ast/v2/translate). In speech-to-text mode it emits source subtitles and translated subtitles as separate events; speech-to-speech mode adds synthesized audio, optionally in a clone of the speaker's voice. It accepts 16 kHz mono WAV only, supports 20 languages plus Cantonese and Shanghainese as sources, and requires Chinese or English on one side.

Side-by-side facts (checked October 17, 2026)

What 1,000 hours cost

Live Interpretation charges about 6.25 tokens per second of input audio, so an hour is roughly 22,500 input tokens. Text output depends on how much people say; the estimate below assumes 15,000 text tokens per hour. Your bill will show the real count in each session's usage event.

The streaming-ASR rows are list prices. The interpretation rows are calculated from the token rates and an assumed text volume, so treat them as estimates. A 1-billion-token prepaid pack lowers the Volcengine token rate to ¥56 per million, but it costs ¥56,000 up front. For a wider grid across vendors, see what 1,000 hours cost once translation, speakers and summaries are added.

When ByteDance is the better pick

• Same-language captions at volume. At $0.14–0.15 per hour, Doubao streaming ASR 2.0 costs well under half of LecSync's rate.
• Mainland Chinese audio and dialects. The default model covers Mandarin, English and eight dialect groups without switching, and the 50-channel default in China is generous.
• Context-heavy recognition. ASR 2.0 accepts dialogue history, hot words and even an image as context, which helps with product names and on-screen terms.
• Dubbed output. Speech-to-speech interpretation with voice cloning is something LecSync does not offer.
• Concurrency-priced plans. The China "concurrency edition" bills ¥500 per channel per month with no hourly meter, which suits always-on lines.

When LecSync fits better

• Pairs that don't involve Chinese or English. Korean to Japanese, Spanish to German, Arabic to French: LecSync handles any pair among 60 languages directly. More on why that matters in built-in translation vs a DIY STT + MT pipeline.
• One stream, one bill. Transcript and translation share a connection and a price. There is no token estimate to reconcile at month end.
• Speaker labels on live captions. Every final segment can carry a speaker with startMs and endMs. See building a meeting client with speakers and timestamps.
• Minutes before the meeting ends. digest: true pushes confirmed sections over the same socket; ByteDance's notes product (Doubao Miaoji) works on recorded files.
• Latency that recovers. Backlog is capped at 2 seconds after a stall, as described in how LecSync caps backlog at 2 seconds.

FAQ

Is Volcengine the same as BytePlus?

Both are ByteDance cloud brands serving the same Doubao/Seed speech models. Volcengine targets mainland China, bills in yuan and uses Beijing endpoints. BytePlus serves international customers, bills in dollars and exposes the speech APIs from its Singapore region. Prices, defaults and feature tables differ slightly between the two.

Does Doubao streaming ASR include translation?

No. Streaming ASR returns text in the spoken language. Translated subtitles come from the separate Live Interpretation model, which is billed by tokens and requires Chinese or English on one side of the pair. LecSync includes translation on the same transcription stream at the same per-minute rate.

Can Doubao label speakers in real time?

Partly. The China streaming docs offer speaker clustering on the Chinese–English model when the second pass is enabled. The BytePlus ASR 2.0 feature table lists diarization for audio files only. Live Interpretation sends a speaker-change flag without names or numbers. LecSync labels every final segment.

How much does an hour of LecSync cost?

$0.36. LecSync bills $0.006 per minute of accepted audio, down to fractions of a second, and that price 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

Choose Doubao streaming ASR when you need same-language captions at the lowest possible hourly price, especially for Mandarin or dialect audio, or when you want dubbed speech-to-speech output. Choose LecSync when your product needs bilingual captions across arbitrary language pairs, speaker labels and running minutes on a single per-minute bill. You can test both claims in an afternoon: start a short stream from the developers page with POST /v1/realtime/connect.