zai-tts

Generate spoken-audio output from text using the GLM-TTS service, driven through the uvx zai-tts command. It fits requests that ask for voice or audio output (often via a "tts" trigger), situations where content is better heard than read — accessibility, multitasking, podcasts, driving, or cooking — and cases that reuse pre-cloned voices.

Before first use, the environment variables ZAI_AUDIO_USERID and ZAI_AUDIO_TOKEN must be set; the documentation explains these are obtained by logging into audio.z.ai and reading localStorage['auth-storage'] from the browser's F12 developer console. Once configured, the basic invocation takes either an inline string with -t or a text file with -f and writes a WAV file to a chosen output path with -o.

Optional flags adjust the delivery: --speed changes playback rate and --volume changes loudness, and --voice selects a speaker. Three built-in system voices are documented — system_001 (Lila, a cheerful standard-pronunciation female voice), system_002 (Chloe, a gentle, elegant, intelligent female voice), and system_003 (Ethan, a sunny standard-pronunciation male voice). The full list of available voices can be retrieved with -l, and users who want a custom voice must first complete voice cloning on the audio.z.ai website before referencing it. As a thin, well-scoped wrapper around the command-line tool, its behavior mirrors whatever the underlying zai-tts binary supports, and outputs are written to the chosen path for playback, sharing, or embedding elsewhere.