Baidu Text-to-Speech

Overview

Baidu Text-to-Speech converts text into speech for applications and devices, with cloud APIs and offline SDKs. It is best suited to teams building Chinese or mixed Chinese-English speech into products, including connected and intermittently connected devices. Its breadth of integration and voice controls is balanced by request constraints and custom pricing for most plans.

Key features

The online service supports Chinese, English, mixed Chinese-English speech, and more than 70 voice styles. Rate, pitch, and volume controls, supported emotional effects, and pronunciation annotations for Chinese polyphonic characters give developers useful ways to shape speech for different contexts. MP3, PCM, and WAV exports cover common audio needs.

Streaming synthesis supports bidirectional text and speech, letting users hear audio nearly as text is entered. Long-text synthesis handles up to 100,000 Chinese characters in a request. Short-text API input must stay under 1,024 GBK bytes, with requests recommended at no more than 120 bytes; that makes it a better fit for brief prompts and announcements than large passages sent through the short-text route.

HTTP and WebSocket APIs sit alongside online and offline SDKs, with Android and iOS SDKs identified for phones and smart devices. Offline synthesis is designed for weak or unavailable network environments, including readers, robots, mobile apps, and smart hardware. The SDK privacy policy describes encryption and safeguards against malicious attacks; Baidu's compliance guide also asks app developers to disclose SDK data collection, methods, and purposes in their privacy policy.

Pricing

Cloud short-text, long-text, and streaming plans use separate voice-library tiers, but all carry custom pricing. Short-text Basic and Premium Voice Library plans each provide 1,000 thousand calls and 100 concurrent requests. The Large-model and Premium tier has the same call allowance but 10 concurrent requests, so it offers less parallel capacity. Long-text plans provide 1,000,000 characters in each of the Basic, Premium, and Large-model and Premium tiers. Streaming plans each provide 1,000,000 characters and 10 concurrent requests. Without prices, these quotas help frame capacity but do not establish which tier is economical.

Offline licensing has clearer terms. The 离线语音合成试用版 costs nothing, lasts 30 days, and permits 2 devices for individual certification or 5 for enterprise certification. It is a limited evaluation rather than a continuing free plan. The 离线语音合成基础版 costs 0.60 CNY, paid once, billed at 0.6元/个, with a one-year license, a 100,000-device minimum, four basic voice libraries, and purchase by device count. The 离线语音合成精品版 costs 0.80 CNY, paid once, billed at 0.8元/个, with the same term and minimum, and includes four basic plus 12 premium voice libraries. Those volume minimums make both paid offline editions a poor fit for small deployments.

Platforms

Baidu lists Android, iOS, Linux, web, API, and self-hosted platforms. The combination of mobile SDKs, API access, and offline synthesis supports both app integration and device deployments, while the intended offline use is particularly relevant when network access cannot be relied on.

Who it's for

Product teams that need Chinese or English speech, adjustable delivery, streaming, or offline playback have a strong set of integration options. The stated use cases—such as audiobooks, news and order announcements, and smart hardware—span short prompts and longer reading. Teams should match the route to the job: the short-text byte limit is restrictive for longer material, while long-text and streaming have separate character-based plans. Developers embedding the SDK should also account for the requested privacy-policy disclosures.

Pros and cons

Pros: Online, offline, HTTP, and WebSocket routes give developers flexibility across connected services and device software.

Pros: More than 70 voice styles, speech controls, emotional effects, and pronunciation annotation provide meaningful control over Chinese speech output.

Pros: Streaming and long-text synthesis address both near-real-time playback and large Chinese passages.

Cons: Short-text requests must remain under 1,024 GBK bytes, and the recommended 120-byte size is a tight constraint for anything beyond brief utterances.

Cons: Most cloud plans have custom pricing, making cost comparison difficult before a workload and tier are selected.

Cons: The offline paid editions require at least 100,000 devices and have one-year licenses, limiting their fit to large deployments.

Alternatives

Text-to-speech tools is a useful place to compare options across the category. Consider Google Cloud Text-to-Speech if a published character rate and free allowance matter: its Standard voices plan is US$0.000004 per character (US$4 per 1 million characters), with 0 to 4 million characters free. Resemble AI is another paid option with a free Flex plan, pay-as-you-go credits, and credits that never expire. Typecast may suit a user seeking a freemium service with a Free plan that includes three projects and 3,000 lifetime credits. TTSReader is a freemium alternative with a free plan. iSpeech offers a free personal-use Basic plan and a Plus plan at $2.95 a month. Oracle Cloud Infrastructure Speech is a freemium alternative with a free monthly transcription allowance. Voicebox is another freemium option, though it does not offer a free trial. Tencent Cloud Text to Speech is a paid alternative with free-plan availability and large-model character bundles.

Verdict

Choose Baidu Text-to-Speech when your product needs Chinese-focused speech controls, multiple integration routes, streaming, or offline synthesis. Its device-oriented offline licensing and broad SDK support are compelling for substantial deployments, but the high device minimums and custom-priced cloud tiers make it harder to recommend for small projects or buyers who need predictable costs before committing.