ASR

Read the audio file with fs, convert it to base64, and pass it to zai.audio.asr.create from the z-ai-web-dev-sdk. The response text field contains the transcription. For quick tasks, run z-ai asr --file ./audio.wav from the CLI instead.

The ASR service supports WAV, MP3, M4A, FLAC, and OGG formats. For best accuracy, use a sample rate of 16kHz or higher, minimize background noise, and keep individual files under 100MB.

No, the z-ai-web-dev-sdk must be used in backend code only and never imported into client-side code. Expose transcription through a server endpoint such as an Express.js route that accepts uploaded audio files.

Empty or incorrect transcriptions usually result from poor audio quality, unsupported formats, or unclear speech. Verify the file format is supported, check that the audio contains clear speech with minimal background noise, and confirm the base64 encoding is correct.

Loop through the file paths, convert each to base64, and call zai.audio.asr.create for each one, collecting per-file success or error results. You can also transcribe an entire directory and write the combined results to a JSON file.