The protocol an iFlytek tutorial targets matters. Its published WebSocket synthesis route differs from the earlier HTTP web API and from a local offline voice engine.
Read the migration notice
iFlytek documents the newer streaming WebSocket route and says the older generic web API is no longer open for new use, while existing users may still have access. New projects should implement the supported current interface, not copy legacy authentication and format assumptions from historical snippets. Record the endpoint, service package and console entitlement on every production deployment.
Respect request sizing and voice activation
The TTS guide specifies a request length in bytes rather than English characters. UTF-8 text with Chinese or other multibyte scripts can hit the byte limit earlier than a character-count estimate suggests. Split scripts at paragraph boundaries with stable IDs. Some languages and voices require activation in the console; a vendor marketing language list is not evidence that a given application already has rights to call a specific voice. Keep a fallback voice for every required locale.
Streaming is not synonymous with incremental input
The documented output may arrive in chunks via WebSocket, but the text payload is submitted in one request with a final data status. Do not design an agent expecting to continually append arbitrary words mid-synthesis without a specifically supported interface. Decode returned base64 audio, order frames correctly and watch terminal flags. The reference lists PCM, MP3 and selected compressed formats and 8/16 kHz output options, so audio downstream may need explicit resampling.
Protect credentials and train reviewers
The published handshake uses an HMAC-based signature with app keys. Generate it on a trusted server and monitor expiry and errors. For education voiceovers, assign native-language reviewers to numerals, mixed-language terms and proper names. A speech file can be mechanically valid while reading a name incorrectly. Keep the original script, generated audio and voice specification for corrections.
Rights and pricing are separate gates
Availability of a voice in a trial or a console menu is not permission to recreate a real person’s identity. Where a custom speaker is involved, collect explicit rights and scope of use. Cost units, voice-package purchase requirements, region access and service limits must be checked with the dated AI10 ledger; this article does not infer subscription or API pricing.
Evidence, availability and commercial-rights boundary
Research cutoff: 25 October 2026. Provider documentation is evidence of its published interface or policy, not independent product testing. No performance score, hands-on trial, customer consent, territory-wide entitlement or legal clearance is claimed. Confirm the exact version, selected region, voice rights, data flow, account limits and current service terms before deployment. Unknown model release dates, account prices and plan limits are deliberately left unasserted; monetary claims belong to the separately dated AI10 pricing ledger. Where documentation is undated or dynamic, the public page is not treated as an archival snapshot of every earlier tier.