Long-form narration calls for job orchestration, not a live-agent audio socket. Tencent’s September 2026 VOD interface provides an explicit asynchronous synthesis model.
Understand the service boundary
TextToSpeechAsync belongs to the video-on-demand API family, not the real-time ASR product. The published interface accepts text for a synthesis task and specifies voice and delivery parameters including speed, pitch, volume, sample rate and output format. The documented maximum text amount is large enough for long material, but does not imply an unlimited uninterrupted human-like performance or zero waiting time. Treat it as a queued job and validate the finished audio before distribution.
A safe long-document production sequence
Split chapters at editorial sections if consistency review or retriable tasks require it. Record a stable task identifier alongside script revision, selected voice, locale, output profile and consent documentation. Generate and retrieve outputs with authenticated server-side requests. Verify that task results correspond to the expected chapter, then listen at joins, pronunciation-sensitive passages and numerical tables. Do not automatically stitch mismatched formats or different voice revisions into a single soundtrack.
Sample rate is not perceived fidelity
An output setting can request a specified sample rate, but it cannot repair errors in the voice model’s pronunciation or pacing. Export at a rate compatible with your video editor or podcast chain, retain an uncompressed master where permitted, then encode a listener delivery version. Apply loudness normalization only after comparing chapters. Be clear to readers that bit rate, sample rate and listening quality are separate considerations.
Billing and account eligibility
The VOD API’s limits, task concurrency and permission model can differ from general Tencent speech-service quotas. A long-text character limit is not a cost estimate. AI 10 must own the canonical price entry, with currency, region, subscription relationship and effective date. Confirm commercial rights for both the script and chosen voice, and obtain consent before attempting identifiable voice replication; standard synthetic voice access does not grant rights to impersonate an individual.
Alternative production routes
For interactive assistants, choose a streaming TTS interface rather than waiting for an entire long-text task. For offline narration, a locally licensed voice model may simplify privacy but usually increases maintenance. For professionally distributed audiobooks, actor narration or detailed post-editing may still be necessary to meet performance expectations. An asynchronous API is a useful production component, not automatic editorial QA.
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.