Home / Models / Speech and vectors

Speech and vectors

Speech synthesis, voice cloning, speech-to-text and text embeddings — sharing the same API key and the same balance as the chat models, with nothing extra to enable.

OpenAI-compatible Tested and working Pay as you go

Four capabilities

Each capability has its own endpoint and billing basis; the prices below are in platform points (¥1 = 1000 points).

Speech synthesis 1 point / char

Text to speech with three voices to choose from: Standard narration, Natural conversation, Lively. Up to 2000 characters per call; output is 22.05 kHz mono WAV.
POST /v1/audio/speech Parameters text / voice / model

Voice cloning 3 points / char

Zero-shot cloning — all you need is one reference audio clip and its matching text, and you can synthesize new content in that voice with no training.
POST /v1/audio/clone Parameters text / prompt_text / prompt_audio_b64

Speech to text 5 points / second

Chinese and English speech recognition, billed by audio duration. Audio is submitted as base64.
POST /v1/audio/transcriptions Parameters audio_b64

Embeddings 3 points / 1k tokens

Text vectorization for semantic search, clustering and RAG. Two models to choose from: bge-m3 (1024 dimensions) and nomic (768 dimensions).
POST /v1/embeddings Parameters input / model

Model list

Speech synthesis

3 voices

Standard narrationtts-001

Natural conversationtts-longxiaoxia

Livelytts-xiaohe

Live

Voice cloning

Zero-shot

Voice cloningvoice-clone

Live

Speech to text

Chinese & English

Speech to text · zh/enasr-zh

Live

Embeddings

Semantic search · RAG

bge-m3 · 1024 dimsbge-m3:latest

nomic · 768 dimsnomic-embed-text

Live

Example calls

The endpoints follow the OpenAI-compatible format, so you can use the official SDKs and just point base_url at ShinLinkAI.

Speech synthesis (curl)

curl https://api.shinloo.com/v1/audio/speech \
  -H "Authorization: Bearer sk-sl-…" \
  -H "Content-Type: application/json" \
  -d '{{"text": "Hello, welcome to ShinLinkAI", "voice": "001"}}' \
  --output hello.wav

Embeddings (curl)

curl https://api.shinloo.com/v1/embeddings \
  -H "Authorization: Bearer sk-sl-…" \
  -H "Content-Type: application/json" \
  -d '{{"model": "bge-m3:latest", "input": "the text to index"}}'

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(base_url="https://api.shinloo.com/v1", api_key="sk-sl-...")

with client.audio.speech.with_streaming_response.create(
    model="tts-001", voice="001", input="Hello"
) as r:
    r.stream_to_file("hello.wav")

Speech synthesis passes the text in the text field (not the OpenAI input); embeddings use the input field. Transcription and cloning require the audio to be submitted as base64.

Read the full docs All models Sign up · Get ¥10

Billing

Units

¥1 = 1000 points. Speech synthesis 1 point/char (¥0.001/char), voice cloning 3 points/char, transcription 5 points/second (¥0.005/second), embeddings 3 points/1k tokens.

When points are charged

Points are charged when the request is accepted; if the upstream fails they are refunded to your balance automatically, so you never pay for nothing.

Rate limits

Audio endpoints: 20 requests/minute; a single synthesis call is capped at 2000 characters.

Shared balance

Shares exactly the same Key and balance as the chat models — no separate top-up.

Live models and prices can be verified with GET /v1/models .