Category: provider

Model Studio Qwen TTS Realtime

Use realtime TTS models for low-latency streaming speech output.

Critical model names

Use one of these exact model strings:

• qwen3-tts-flash-realtime
• qwen3-tts-instruct-flash-realtime
• qwen3-tts-instruct-flash-realtime-2026-01-22
• qwen3-tts-vd-realtime-2026-01-15
• qwen3-tts-vc-realtime-2026-01-15

Prerequisites

• Install SDK in a virtual environment:

python3 -m venv .venv
. .venv/bin/activate
python -m pip install dashscope

• Set DASHSCOPE_API_KEY in your environment, or add dashscope_api_key to ~/.alibabacloud/credentials.

Normalized interface (tts.realtime)

Request

• text (string, required)
• voice (string, required)
• instruction (string, optional)
• sample_rate (int, optional)

Response

• audio_base64_pcm_chunks (array)
• sample_rate (int)
• finish_reason (string)

Operational guidance

• Use websocket or streaming endpoint for realtime mode.
• Keep each utterance short for lower latency.
• For instruction models, keep instruction explicit and concise.
• Some SDK/runtime combinations may reject realtime model calls over MultiModalConversation; use the probe script below to verify compatibility.

Local demo script

Use the probe script to verify realtime compatibility in your current SDK/runtime, and optionally fallback to a non-realtime model for immediate output:

.venv/bin/python skills/ai/audio/alicloud-ai-audio-tts-realtime/scripts/realtime_tts_demo.py \
  --text "这是一个 realtime 语音演示。" \
  --fallback \
  --output output/ai-audio-tts-realtime/audio/fallback-demo.wav

Strict mode (for CI / gating):

.venv/bin/python skills/ai/audio/alicloud-ai-audio-tts-realtime/scripts/realtime_tts_demo.py \
  --text "realtime health check" \
  --strict

Output location

• Default output: output/ai-audio-tts-realtime/audio/
• Override base dir with OUTPUT_DIR.

References

• references/sources.md