Billing items

Free trial vs. Commercial Edition

All four services offer a three-month free trial. Upgrade to Commercial Edition when you need higher concurrency or longer daily recording limits.

Pay-as-you-go rates

Usage is billed daily. All usage within a calendar day is priced at the single tiered rate that corresponds to the total daily volume — not split across tiers.

Bills are settled daily at 24:00 (UTC+8). Today's usage appears on the next day's bill. Keep your account balance sufficient to avoid service interruptions.

Real-time speech recognition — Standard rate: USD 1.40/hour

Billing is based on actual audio streaming duration, measured in seconds (rounded down). For example, 22.8 seconds of audio is billed as 22 seconds.

Billing starts when the StartTranscription command is sent and audio streaming begins — not when the WebSocket connection is established. If a call exceeds the maximum session duration limit, it fails without incurring a charge.

Short sentence recognition — Standard rate: USD 1.40/thousand calls

Failed calls are not counted.

Recording file recognition — Standard rate: USD 1.00/hour

Billed per second, rounded down. Upgrade to Commercial Edition if you need more than 2 concurrent calls or more than 2 hours of recording recognition per calendar day.

Speech synthesis — Standard rate: USD 1.40/thousand calls

Speech synthesis is billed by the number of UTF-8 encoded characters per request, not by audio output duration. Each Chinese character, English letter, or punctuation mark (full-width or half-width) counts as one character. SSML tags are excluded. The maximum is 300 characters per request. Failed calls are not counted.

You can check your actual call volume on the Statistics page of the ISI console to verify your bill.

Billing example

All usage within a calendar day is priced at the single tiered rate for the total daily volume.

Self-learning platform

The self-learning platform is free. Create up to 10 custom language models to improve recognition accuracy for business-specific vocabulary and phrases.

Speech synthesis speaker customization

Pricing varies based on business scenario, data volume, speaker type, and intellectual property rights. No standard rate applies.

To get a quote, contact nls_support@service.aliyun.com.

CosyVoice version and channel billing differences

Q: Why does CosyVoice 3-plus cost more than CosyVoice flash?

CosyVoice pricing reflects voice cloning capability, synthesis quality, and feature scope — not just the number of built-in voices. CosyVoice 3-plus offers higher-quality synthesis and more advanced features than flash, which is reflected in its price.

Q: Is the price difference between CosyVoice 3.5 and CosyVoice 3 a permanent pricing structure?

Yes. The price difference between CosyVoice 3.5 and CosyVoice 3 reflects their respective capabilities and is part of the standard pricing structure — not a limited-time promotion or temporary discount.

Q: What is the difference between Alibaba Cloud Model Studio CosyVoice and ISI CosyVoice?

Both platforms bill by the number of characters. Key differences:

• Alibaba Cloud Model Studio CosyVoice: New users receive a free quota of 10,000 characters (valid for 90 days). Standard price is approximately ¥0.003 per 1,000 characters.

• ISI CosyVoice: No free quota for new users. Base TTS price starts at approximately ¥0.004 per 1,000 characters.

Choose the platform that best fits your eligibility and voice quality requirements.

Q: Is the fun-asr-far-field model billed at the same rate as the default model?

Yes. The fun-asr-far-field model is billed at the same commercial rate as the default model. No additional charges apply for using this model.

Offline speech synthesis SDK authorization and resource plan usage

Q: What does the "number of authorizations" mean for the offline speech synthesis SDK?

The authorization count is the maximum number of devices that can activate and run the SDK — not the number of downloads or installations. Each device that initializes the SDK (identified by a unique device_id) consumes one authorization. Reinstalling on the same device does not consume an additional authorization if the device_id is unchanged. The project code, AccessKey, and app_key can be shared across devices.

Q: How do I track the number of activated devices?

No API or console view is available for querying total activated devices. Track activations using the task_id from SDK callbacks combined with your application logs.

Q: Does the offline speech synthesis resource plan support Cantonese?

No. The offline speech synthesis service does not currently support Cantonese.

Q: How does the Long Text-to-Speech (LongTTS) resource plan work?

The resource plan with commodity code FPT_nls_longtts_bag_deadlineAcc_1565165090 applies to the Long Text-to-Speech (LongTTS) service. After purchase, the plan automatically deducts from usage fees — no manual configuration is required.

Resource plan purchase page is inaccessible or appears blank

If the resource plan purchase page fails to load or appears blank, try the following steps:

1. Use a direct purchase link:

   Go directly to Speech Synthesis Resource Plan purchase to purchase speech synthesis resource plans.

2. Clear browser cache: Open Google Chrome, clear the browser cache, then restart the browser and try again.

3. Use the documentation entry: If the direct link is still unavailable, navigate to the resource plan purchase page through the standard entry in the ISI billing documentation.

Services and features that do not incur separate charges

The following ISI services and features do not incur separate charges:

• SDM (Speech Data Management): SDM is free. You are responsible for provisioning and maintaining the deployment environment.

• Alert recipients: Adding and notifying alert recipients is free.

• MRCP Server: MRCP Server has no separate charge. Usage is billed under the standard ISI billing items.

• Recording File Recognition API: No additional activation fee is required. Standard pay-as-you-go rates apply.