Qwen TTS demo

Each sample plays with one click. The emotional Serena English recording lasts 7.52 seconds. The voice-cloning recording also lasts 7.52 seconds and uses the first synthetic sample as its reference. It does not use a real person's recording. The mixed Vivian recording lasts 9.92 seconds and combines English with Mandarin.

These are fixed recordings. Replaying one revisits the same delivery; it does not generate a new performance. Hear all three before deciding which qualities matter most for your task. For background after listening, open the model guide.

Use the same headphones or speakers at a comfortable, consistent volume. First, follow the meaning. Then, focus on individual words. Finally, listen without reading along: a passage can seem clearer when the text supplies clues your ears missed.

• Pronunciation: are consonants, word endings, and syllables distinct enough to understand?
• Pacing: do important words have room, or does the delivery rush through them?
• Pauses: do breaks support the sentence structure and leave related words together?
• Consistency: does the voice stay recognizable through the beginning, middle, and ending?

A short demo cannot cover every surname, abbreviation, date, or technical term. Notice what its text demonstrates, then list the gaps relevant to your script. A clear everyday sentence says little about a specialized product name.

For your later generation test, include one difficult item at a time. Try a date or a price in a sentence whose intended reading is unambiguous. If the result differs from your intention, spell out the spoken form. Compare a comma with a full stop while leaving the other words unchanged; punctuation is worth testing, without assuming an exact pause duration.

The Vivian sample gives you a place to hear English and Mandarin in one recording. Listen to the final word before each language change and the first word after it. Check whether the transition feels abrupt, whether either phrase accelerates, and whether both languages remain easy to follow.

Use Serena's English sample as another listening reference, remembering that the voice and text differ. This prevents a controlled language comparison. In your own bilingual test, choose a familiar English phrase and a short Mandarin clause. Ask a fluent listener to assess the language you know less well instead of judging it solely by a smooth sound.

Listen for similarities in voice color and rhythm between the first synthetic recording and the cloning sample. Similarity and intelligibility are separate questions: a voice can sound familiar while a word still needs checking. Make a note of both instead of reducing the comparison to a single impression.

This synthetic reference example cannot predict how your microphone, room noise, or speaking style will affect a personal reference. For reference-based cloning, choose audio you are entitled to use and observe the workspace's 2 MB reference limit.

qwentts.app is an independent browser workspace. When you are ready to generate custom speech, open the workspace and log in. A signed-in free account has one generation per day with up to 120 characters per request. Paid requests allow up to 1,000 characters. Listening to these public recordings does not spend that daily generation.

Pick a preset voice and an appropriate language. Keep the first test short enough to inspect every word, and save the WAV download for comparison. Record your input and planned changes; with a daily allowance, one focused test is easier to learn from than a sentence crowded with unrelated challenges.