Terms

Audio sampling rate

The audio sampling rate is the number of times per second that a recording device samples a sound signal. A higher sampling rate produces more accurate and natural sound reproduction.

Voice Service supports two sampling rates: 16000 Hz and 8000 Hz. Telephone services typically use 8000 Hz. All other services use 16000 Hz.

When calling Voice Service, convert audio data with a sampling rate higher than 16000 Hz to 16000 Hz before sending it to the service. If your audio data uses 8000 Hz, do not convert it to 16000 Hz. Instead, select a model in your project that supports 8000 Hz.

Audio bit depth

Audio bit depth measures the amplitude resolution of each sample. It reflects how precisely the sound waveform is captured—or the resolution of the sound card. Higher values mean higher resolution and better sound fidelity.

In Voice Service, the most common audio bit depth is 16-bit little-endian. Each sample uses 2 bytes of storage. That means 2 bytes store 1/16000 second of audio data.

Each sample records amplitude. Sampling precision depends on bit depth:

• 1 byte (8 bits) records 256 amplitude levels—dividing amplitude into 256 steps.

• Two bytes (16 bits) represent 65,536 values.

A 2-byte bit depth meets CD-quality standards.

Audio encoding format

An audio encoding format defines how audio data is stored and transmitted. Note that an audio encoding format differs from an audio file format. For example, a WAV file includes a header that specifies the encoding used for its audio data—such as PCM, AMR, or another format.

Appkey

Each project you create in the Tongyi Tingwu console has a unique identifier called an Appkey. You must provide this Appkey when calling Voice Service. The service uses it to retrieve your project’s configuration.

When multiple business scenarios require Voice Service—such as call center support and mobile keyboard input—each scenario requires different speech capabilities. Best results occur only when your project configuration matches your business scenario.

AccessKey

An AccessKey is a credential for programs to access Alibaba Cloud APIs. Go to the AccessKey management page to create and view your AccessKey.

An AccessKey consists of two parts: an AccessKey ID and an AccessKey secret. The AccessKey ID acts like a username. The AccessKey secret signs your API requests to prevent tampering. Use both together. Treat your AccessKey secret like a logon password—never share it with anyone.

Intermediate result

You can choose whether to return intermediate results when calling Voice Service:

• Set to false: Return one complete result only after full recognition finishes.

• Set to true: Return intermediate results while you speak, plus the final complete result.

For example, if the final recognized text is “Hello Alibaba”, enabling intermediate results returns five partial results as you speak, like this:

You
Hello
Hello there
Hello Alibaba
Hello Alibaba

action_id

Each Voice Service request has a unique action_id, automatically generated by the API. Use it to troubleshoot issues.

Multi-channel audio stream

A special audio stream type used in online meetings. It includes one mixed audio stream plus separate streams for each speaker.

Protobuf

Tongyi Tingwu uses Protocol Buffers, a data description language developed by Google, to describe the encapsulation structure of multi-channel audio stream data during online multi-person meetings.

Chapter overview

Splits audio or video content into segments by topic. Generates a title for each segment and extracts a summary. Helps you quickly grasp content structure and context.

Summary

Uses Qwen Large Language Model (LLM) capabilities to generate full-text summaries, speaker summaries (requires speaker diarization), Q&A recaps, and mind maps. Call these features after transcription completes. Supports Chinese and English.

Key point extraction

Includes four AI capabilities: keyword extraction, to-do item detection, scenario identification, and key content extraction. Runs after transcription completes.

PPT extraction and summary

Extracts slides shown in a video and summarizes the spoken content for each slide. Solves the problem of missing PPT files and links each slide to its corresponding spoken content. This feature combines AI capabilities across three modalities: speech, language, and image.