Aliun released a Qwen-TTS speech synthesis model, close to the real voice, supporting Mandarin, English, and three local Chinese dialects.

Frontier Trends

The core capability is to convert the text entered into Chinese and English into a voice output with natural expression. Compared to the traditional TTS model, the biggest bright spots for Qwen-TTS are:

  • ** High Nature**: Voice expression is closer to the person, with emotion, rhythm, tone change;

  • ** Multilingual and dialectic support**: currently in favour of Mandarin, English and three Chinese dialects (Beijing, Shanghai, Sichuan);

  • ** Multisound selection**: Provides different genders, tone and accents, adapted to diverse scenes.

Words and dialects support the details

Supported speech variant: 真实合成样例：https://qwenlm.github.io/zh/blog/qwen-tts/

Technical principles and data base

  1. Large-scale training materials support The model training used 3 million hours of voice data, including Chinese and English alignment data, as well as a wealth of dialectic material, which enabled the model not only to be natural in voice but also to imitate the language style of different regions.
  2. Rhythm and Emotional Modelling Qwen-TTS supports ** automatic adaptation of text speed, accent, rhythm and emotional performance**. For example, when expressing surprise, tenderness or anger, the voice automatically reflects the corresponding emotion, without the need for an overt sign.
  3. Sound modelling and style migration

  • The model allows the same sentence to produce a voice output in a variety of styles (e.g. male/female, Northern/South).

  • Current support 7 sounds: Cherry, Ethan, Chelisie, Serena (Chinese and English)

  • Dylan, Jada, Sunny.

Performance assessment: 模型在 SeedTTS-Eval 评测集上的指标如下： Core indicators:

  • **WER (word error rate): ** The accuracy of speech recognition reverse text should be as low as possible;

  • SIM (symmetry of sound) measures the degree of proximity between the generation and the target sound, as high as possible.

How to use: API and code access

** Prices:** Qwen-TTS 通过 Qwen API 的方式提供，用户可通过如下 Python 示例调用模型完成语音合成任务： Basic processes: Set the API key (DashScope API); Set synthesizing statements, target sounds and model versions; Call synthetic functions to get audio links; Downloads and saves audio files. Example snippet (simplified): %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %2 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 %1 It’s not like I’m going to have to do it again. #Import dashcope # If get_api_key(): Api_key = os.getenv (“DASHSCOPE_API_KEY”) If not api_key: “DASHSCOPE_API_KEY environmental protection not set.” ♪ Return api_key ♪ If synthesize_speech (text, voice=”Dylan”, model=”qwen-tts-latest”): == sync, corrected by elderman == @elder_man Try: = dashscope.audio.qwen_tts.SpeechSynthesizer.cal(

Model=model

I don’t know what you’re talking about, but I don’t know what you’re talking about. I don’t know what you’re talking about, but I don’t know what you’re talking about. It’s not like I’m going to have to go back to school. I’m not sure what I’m talking about. Check if excuse is None If excuse is None: “API called returned None response”

Check if excuse.output is None

If excuse.output is None: I’m sorry, RuntimeError (“API call failed: recall.output is None”)

Check if excuse.output.audio exists

If not hasattr (response.output, ‘audio’) or recall.output.audio is None: “API calls failed: recall.output.audio is None or missing” Audio_url = recall.output.audio [“url”] ♪ Turn audio_url ♪ Exception as e: “Speech synthesis failed: {e}) If you don’t like it, you’ll have to take care of it. Try: Resp = requests.get (audio_url, timeout=10) I don’t know what you’re talking about, resp.raise_for_status() With open (save_path, ‘wb’) as f: F.Write (resp.content) I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, but I’m sorry, I’m sorry, I’m sorry, I’m sorry, but I’m sorry, I’m sorry, I’m sorry, I’m sorry, I’m sorry, but I’m sorry, I’m sorry, I’m sorry, I’m sorry, I’m sorry, I’m sorry, I’m sorry, I’m sorry, but… Exception as e: “Download failed: {e}) Def main(): Text = ( Yo, guess what? I’m looking at the NBA today, and Curry’s pitching and playing, and I’m gonna have to call him “Father” in the basket. I’m not sure what I’m talking about. “downloaded_audio.wav” Try: Audio_url = synthesize_speech(text) I don’t know what you’re talking about, but I don’t know what you’re talking about. Exception as e: print(e) If name_ = “main”: Main()

← Previous Google releases Gemma 3n-a new generation of lightweight multimodule AI models with a response speed increase of about 1.5 times 2GB memory cell phones running.
Next → Aliun publishes a multi-model model of Qwen V Lo, which produces images and edits images in natural languages.

Related Posts

No related posts found