GLM TTS | EmpirioLabs AI Docs

Z.ai · Audio Generation

POST /v1/audio/speech

LLM-based text-to-speech with zero-shot voice cloning from 3-10s of audio and emotion-expressive, controllable output via multi-reward RL.

At a glance

Field	Value
Model id	`glm-tts`
Input modalities	Text, Audio
Output modalities	Audio
Context window	-
Weight precision	INT8 / FP16
Features	voice_cloning, emotion_control
Native inference	Yes
New	No
Supported endpoints	`POST /v1/audio/speech`

Pricing

Charge	Spec	Rate
Fast (INT8)	per 1k characters	$0.20
Quality (FP16)	per 1k characters	$0.21

Example request

$ curl https://api.empiriolabs.ai/v1/audio/speech \
>   -H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \
>   -H 'Content-Type: application/json' \
>   -d '{"model": "glm-tts", "input": "Hello from EmpirioLabs."}'

Parameters

Parameter	Type	Required	Default	Description
`input`	string	yes	-	Text to synthesize. For multi-speaker use [S1] / [S2] tags or ‘Speaker N:’ lines.
`voice`	enum	no	`"emma"`	emma=English Female, james=US Male, arthur=US Male alt, xiaomei=Chinese Female, zhigang=Chinese Male, custom=upload reference via voice_audio_url. · Allowed: `emma`, `james`, `arthur`, `xiaomei`, `zhigang`, `custom`
`voice_audio_url`	string	no	-	Reference audio URL for custom voice cloning. The reference recording must contain the speaker reading this exact consent phrase aloud, in their own voice: “I consent to Empirio Labs cloning my voice for the purpose of generating synthetic speech. I understand that my voice sample will be used to create personalized audio content.” Reference audio without the phrase is rejected.
`output_format`	enum	no	`"mp3"`	Output media file format (mp3, wav, mp4, png, jpg, etc., depending on the endpoint). · Allowed: `mp3`, `wav`
`speed`	number	no	`1.0`	Speaking rate multiplier. · Range: 0.5 – 2.0
`model_quality`	enum	no	`"quality"`	quality=FP16 (better), fast=INT8 (quicker) · Allowed: `quality`, `fast`
`sample_rate`	enum	no	`"24000"`	Output sample rate in Hz. · Allowed: `24000`, `16000`
`volume`	number	no	`1.0`	Output gain multiplier. · Range: 0.1 – 2.0
`use_cache`	boolean	no	true	Speeds up repeated identical generations.
`optimize_input`	boolean	no	true	Auto-fix pronunciation of technical terms, acronyms, and special characters.
`seed`	number	no	-	Reproducibility seed.

Notes

Limits

Max input: 5,000 characters
Generation: 5-10 minutes

Voice cloning

Reference audio: 3-10 seconds
Accepted formats: WAV, MP3, OGG, FLAC, AAC, M4A, WebM

Preset voices

emma (English F)
james (US M)
arthur (UK M)
xiaomei (Chinese F)
zhigang (Chinese M)

Machine-readable schema: GET https://api.empiriolabs.ai/v1/models/glm-tts.