Baidu Cloud Speech Synthesis

In one sentence: the site's cheapest tier lives here, and it's the only brand where reading English is clearly cheaper than reading Chinese. The trade-off is a billing method unlike anyone else's, and short sentences look expensive.

---

What This Is

Baidu Cloud's online speech synthesis (short text). Accurate, stable Chinese pronunciation at an extremely low price.

Like Tencent Cloud, Baidu has only one "model" option in Soniva — the three price tiers are distinguished by the voices themselves.

---

The Three Tiers & Pricing

Price source: https://cloud.baidu.com/doc/SPEECH/s/Ql9misjot (verified 2026-08-18)

Note even the most expensive tier is only 12 credits/1,000 chars — Baidu is cheap across the board. The gap between tiers (2.4×) is far smaller than Tencent's 21×, so you can just pick the best-quality Ultimate tier inside Baidu without regret. The free-preview representative voice, Du Hanzhu, is on the LLM Voice Library tier.

---

Billing: Per Call, Not Per Character

This is Baidu's biggest difference from other brands, and the only part you really need to understand.

Baidu's official rule:

Every 120 GBK bytes counts as 1 call; anything under 120 bytes also counts as 1 (rounded up).

Under GBK encoding: 1 CJK character or Chinese punctuation mark = 2 bytes; 1 English letter/digit = 1 byte. Converted:

This Produces Two Practical Consequences

1. Reading English costs half of Chinese. For the same 1,000 characters, Chinese takes 2,000 bytes (17 calls), while English takes only 1,000 bytes (9 calls). The credit figures in the table above are calculated for pure Chinese; when you read English, the actual deduction is roughly half. Soniva's deduction logic computes by real byte count precisely — it's not an estimate.

2. Very short content looks more expensive per unit. Because of rounding up, even reading a single character counts as 1 call. Soniva specifies that any real call deducts at least 1 credit (to avoid rounding down to 0 and free-riding), so a single sentence converts to a higher "per-1,000-char unit price" than the table. That's inherent to per-call billing, not an overcharge.

The longer you read, the closer the actual unit price gets to the table's figure.

---

Voices

Baidu doesn't expose a public voice-list interface, so Soniva uses a built-in static voice table, cross-checked against the official docs, covering the Basic / Premium / Ultimate / LLM four groups.

Most LLM Voice Library voices support multi-emotion expression (neutral, happy, sad, angry, surprised, fearful, etc.); the voice's description states exactly which emotions it supports.

---

When to Choose Baidu

---

Related Documents

• Full Credit Rate Table (Detail 2 explains exactly this per-call billing rule)
• What a Given Credit Budget Buys
• Eight Brands at a Glance