Publish a Model (Speech)

Target Outcome

The speech model passes protocol testing, is published to the intended scope, and returns a playable or decodable audio result.

Applicable Roles

• Model Provider

Before You Start

• Prepare the model source, identifier, API credential, endpoint, and a non-sensitive audio or text sample.
• Confirm language, voice, format, sample rate, synchronous or asynchronous behavior, billing, and limits.

Procedure

1. From the platform home page, select My Models in the left navigation.
2. Open My Publications. Use Public Models / Private Models to switch publication areas, or open Overview and My Aggregations when needed.
3. Select Publish Model in the upper-right corner.
4. Select a publication area:
   • Publish to Private Area makes the model visible only within the team or tenant and keeps it out of the public catalog.
   • Publish to Public Area lists the model in the public catalog for all tenants and allows independent pricing and rate limits.
5. Select Publish to Public Area to open Step 1.

Step 1: Basic Information

• Under Model Source / Meta-Model Information:
  • Select a meta-model, such as qwen3-tts-flash.
  • Select a model source, such as Alibaba - China.
  • Enter the request URL, such as https://dashscope.aliyuncs.com.
  • Enter the API key in the protected field, such as sk-***.
  • Enter the exact upstream Model Source ID, such as qwen3-tts-flash.

• Confirm Speech Model and select the correct subtype, such as Text to Speech.

• Under Request Headers, keep the default Authorization: Bearer <key> template and add only headers required by the upstream service.

• Under Model Parameters, set the input modality to Text and the output modality to Audio.

• Under Supported Protocols and Default Parameters:
  • Select OpenAI-Audio, run the connectivity test, and enter the endpoint.
  • Configure inputs such as Text, Voice, Language, Audio Format, Sample Rate, Volume, Speech Rate, Pitch, and Seed.
  • Select Synchronous or Asynchronous invocation.
  • Configure result parsing with Result Path, URL Extract Field, and Base64 Extract Field.

• Enter the public Custom Identifier and description.

• Select Publish Immediately or Scheduled Publication.

• Select Next to open Step 2.

Step 2: Billing Configuration

• Select Character Billing or Free.
• For character billing:
  • Enable Show Price Comparison when a reference price should be displayed.
  • Enter the input sale price and optional original price in Credits per 1M characters.
  • Optionally configure a free quota, eligible-user count, and total amount.

• Select Next to open Step 3.

Step 3: Rate-Limit Configuration

• Select Enable Rate Limiting or Disabled.
• Configure default RPM and TPM values, or set either limit to Unlimited.

• Select Save Only or Submit for Review.

Parameter Reference - Speech Model

Field	Type	Example	Description
Meta-Model	Select	qwen3-tts-flash	Required; base meta-model
Model Source	Select	Alibaba - China	Required; upstream Model Provider
Request URL	URL	https://dashscope.aliyuncs.com	Required; model-service base URL
API Key	Password	sk-***	Required; protected upstream credential
Model Source ID	Text	qwen3-tts-flash	Required; exact upstream model name
Model Type	Single select	Speech Model	Required; model function
Model Subtype	Select	Text to Speech	Required; speech-model subtype
Request Headers	Key-value pairs	Authorization: Bearer <key>	Optional; authentication and custom headers
Input Modality	Multi-select	Text	Required; accepted input type
Output Modality	Multi-select	Audio	Required; result type
Supported Protocol	Multi-select	OpenAI-Audio	Required; test connectivity before continuing
Endpoint	URL	https://dashscope.aliyuncs.com/api/v1/services/audio/tts/SpeechSynthesizer	Required; protocol endpoint
Input Parameters	Parameter list	Text / Voice / Language / Audio Format / Sample Rate / Volume / Speech Rate / Pitch / Seed	Optional; protocol inputs and required-state settings
Invocation Method	Single select	Synchronous / Asynchronous	Required; invocation behavior
Result Path	Text	data.audio or response.output.results	Optional; path to the result payload
URL Extract Field	Text	url or audio_url	Optional; field containing the result URL
Base64 Extract Field	Text	b64_audio	Optional; field containing Base64 audio data
Custom Identifier	Text	qwen3-tts-flash	Required; model identifier shown to users
Description	Text	Text to speech...	Optional; model description
Publication Method	Single select	Immediate / Scheduled	Required; publication time
Billing Method	Single select	Per Character / Free	Required; billing method
Show Price Comparison	Switch	On / Off	Optional; displays an original reference price
Input Sale Price	Number	8 Credits/1M characters	Required for paid models
Original Price	Number	16 Credits/1M characters	Optional; character reference price
Free Quota	Switch	On / Off	Optional; configures free usage quota
Rate Limiting	Single select	Enabled / Disabled	Optional; controls invocation limits
RPM	Number / Unlimited	2 requests/minute	Optional; request limit per minute
TPM	Number / Unlimited	100 tokens/minute	Optional; token limit per minute

Completion Checklist

Purpose: These are the exit criteria for the current feature task. Use them to decide whether the result is observable and reviewable and whether you can continue to the next step in the scenario. They do not repeat the procedure; if any item fails, follow the troubleshooting section below.

Check	Pass Criteria
1	Protocol connectivity passes and language, voice, and format settings are accurate.
2	Publication or review status is correct.
3	A controlled call returns playable audio and the call log is traceable.

Troubleshooting

Symptom	Check First
Protocol test fails	Endpoint, credential, model identifier, audio encoding, and request body
Audio cannot be played	Response mapping, Content-Type, sample rate, format, and result URL

User Manual

Review complete My Models fields and publication-result validation