Skip to main content

Qwen-Audio 3.0 TTS Flash accepts a script and performance directions for assistant speech, using 597 preset voices with configurable speed, pitch, volume and audio output.

From $0.014/1,000 characters

Workbench

Input

Text to speech
128/10000
3.0 TTS Flash script: required text, up to 10,000 characters, with no default script; its length determines the charge per 1,000 characters. at least 1 characters · at most 10000 characters

128 characters

3.0 TTS Flash casting: choose one of 597 preset voices belonging to this model; other-model IDs are rejected, no default is specified in the supplied facts, and selection does not affect price.
36/1000
3.0 TTS Flash direction: free-style emotion, tone, persona, pace or dialect notes that are not spoken or billed; no length limit or default direction is specified in the supplied facts. at most 1000 characters

Never read aloud. It only shapes how the script is performed.

Safety checkerOff

SpicyAPI does not run a safety checker on this model. Your prompt and media go to the model as written, and the model itself decides what comes back.

Checking your account before enabling submission and pricing.

Output

Ready for a new run
Example result
0:00 / –:––

Related models and next tasks

Missing the model you need?

Practical speech for the next customer reply

Qwen-Audio 3.0 TTS Flash is aimed at assistant and customer-service speech, with low first-packet latency as its stated focus. This SpicyAPI endpoint returns audio through task submission and polling, with 597 preset speakers and a price of $0.01425 per 1,000 text characters. Luna Wang is the default English voice. SpicyAPI imposes no content filter; this model delivered our NSFW test scripts unchanged in the November 5, 2026 trials.

Open the playground
Practical speech for the next customer reply
Service voices
597 presets; no cloning
Voice distribution
589 Chinese-language, 8 English
Default voice
English: Luna Wang
Text limit
10,000 characters per task
Usage price
$0.01425 / 1,000 characters
Result workflow
Submit, then poll the task
Service speech
Built around assistant responsesAlibaba positions 3.0 Flash for low first-packet latency in assistants and customer service. Here, plan the integration around a submitted task and a completed audio result rather than assuming a streaming connection.
Speaker selection
Choose a voice for your serviceThere are 589 Chinese-language presets and eight English presets, with English speaker Luna Wang as the default. Audition the greeting, an answer and a notification before choosing a service voice.
Cost
Count the reply charactersOnly text determines the charge, calculated linearly at $0.01425 per 1,000 characters. Performance instructions are unbilled, and changing the voice or audio settings does not add a surcharge.

What the Qwen-Audio 3.0 TTS Flash API does well

  • Assistant-focused model positioning

    This tier targets low first-packet latency for service dialogue. That positioning does not establish a completion-time guarantee for a task on this endpoint.

  • A voice for routine customer exchanges

    Select from 597 presets to audition an appropriate service speaker. The eight English choices sit alongside 589 Chinese-language voices; this endpoint accepts its own voice IDs only.

  • Clear separation of reply and manner

    Put the customer-facing answer in text and describe the service manner in instructions. Up to 1,000 characters of guidance can request a calm, patient or concise delivery without being spoken.

  • Controls for service phrasing

    Tune speed and pitch within 0.5–2 and loudness within 0–100. For pause, pronunciation or emphasis markup, supply SSML in text and enable its interpretation.

  • Encoding to match the receiving system

    MP3, WAV, raw PCM and Opus are available, with MP3 as the default. Match sample_rate to a supported option from 8–48 kHz; an Opus request can also set bitrate.

  • Language hints for short answers

    Automatic detection is available, together with ten explicit language hints. When a service reply is short, audition it with the intended hint and verify any names or local terms.

Who Qwen-Audio 3.0 TTS Flash is for

  • Turn a prepared answer into speech

    Voice assistant developers

    Turn a prepared answer into speech

    Generate a concise answer to a question such as the next train time or a calendar check. Account for task completion in the product flow and audition how the answer sounds when played.

  • Read a hotline answer clearly

    Customer-service teams

    Read a hotline answer clearly

    Prepare an opening answer, a routing instruction or a standard service explanation. Check the selected preset on menu options and pauses before using the recording in a hotline.

  • Speak the update that matters

    Booking and delivery operators

    Speak the update that matters

    Voice a booking confirmation, pickup reminder or delivery notice. Review the date, time and destination wording so the listener can understand the next action.

  • Keep device prompts brief

    Device and automotive teams

    Keep device prompts brief

    Prepare status messages for smart devices or in-car systems. Select a supported encoding and sample rate for the receiving player, then check audibility and phrasing on that system.

How to call the Qwen-Audio 3.0 TTS Flash API

  1. Prepare the customer-facing words

    Use text for the service answer or notification, up to 10,000 characters. Write dates, times and option names as you want the listener to hear them.

  2. Set the service speaker and language

    Choose a voice_id from the 3.0 Flash enum; leaving it unset uses Luna Wang. Select a language hint if needed, or retain auto detection.

  3. Configure delivery and audio

    Add a brief service-tone note in instructions and adjust pace if necessary. Pick an output format and supported sample rate, using bitrate only with Opus.

  4. Submit and track the task

    Keep the returned task reference and poll for the finished result. Do not treat low first-packet positioning as a streaming API contract.

  5. Check the message on its destination

    Listen to the result in the intended playback system before using it for customers. Copy useful output files within the 14-day retention window.

Qwen-Audio 3.0 TTS Flash worked examples

  • A voice assistant checks a departure

    A concise prepared answer uses the default English preset explicitly, with a neutral service direction.

    {"text": "The next train leaves at ten twenty from platform four. You have twelve minutes to reach the station.", "voice_id": "qwen-audio-3.0-tts-flash-loonglunawang", "language": "en", "instructions": "An assistant giving clear travel information in a calm, practical tone."}
  • A hotline explains opening hours

    A slightly slower reading leaves room for the opening times and the next step to be understood.

    {"text": "Our support team is available from nine in the morning until six in the evening. Please leave your reference number after the message.", "voice_id": "qwen-audio-3.0-tts-flash-loonglunawang", "language": "en", "instructions": "A service agent, courteous and patient, with clear separation between the two sentences.", "speed": 0.9, "output_format": "wav"}
  • A delivery pickup notice

    Opus encoding and an allowed bitrate show how a notification can specify its playback format.

    {"text": "Your parcel is ready for collection at the reception desk. Please bring your collection code before seven this evening.", "voice_id": "qwen-audio-3.0-tts-flash-loonglunawang", "language": "en", "output_format": "opus", "sample_rate": 24000, "bitrate": 64000}

Qwen-Audio 3.0 TTS Flash API questions

Start with the service messages customers will hear most often: a greeting, a routine answer and a notification containing a time or reference. Audition the same preset on all three, paying attention to option names and the information that determines the listener’s next action. Keep delivery notes simple enough to reuse, while placing changing customer-facing wording in text. Confirm the format and sample rate against the destination player before producing a larger set of messages. In a live assistant, account for task submission, polling and playback in the user experience instead of treating the model’s latency positioning as an end-to-end response guarantee.