Voice model · MiniMax
MiniMax Speech 2.6 HD is MiniMax’s high-definition text-to-speech model, released at the end of October 2025. On Deepnia it reads up to 5,000 characters per generation in 23 languages or with auto-detect, at a speed you set from 0.70× to 1.20×. Its strength here: speaking in your own voice, cloned with your consent from a 10–60 second recording.
The voice generator opens with MiniMax Speech 2.6 HD already selected. The exact cost shows before every creation; you sign up when you generate.
In the voice clone tool, record or upload 10 to 60 seconds of clear speech and confirm you have permission to use the voice. It then appears at the top of the MiniMax voice list, with your personal voices. MiniMax says its Fluent LoRA technique keeps a speaker’s timbre even when the recording is accented or hesitant.
MiniMax says Speech 2.6 reads web addresses, emails, phone numbers, dates and amounts without you spelling them out. Handy for a phone greeting, an appointment reminder or a delivery message.
Pick the language of your text from 23, including English, French, Spanish, Arabic, Hindi and Brazilian Portuguese, or let auto-detect decide.
MiniMax designed the HD version of Speech 2.6 for audio quality and similarity to the original voice: an asset for a polished voice-over and for your cloned voice.
Set the speed from 0.70× to 1.20× in 0.01 steps. Go slower for a number or opening hours people need to note, faster for an upbeat announcement.
In the voice clone tool, record or upload 10–60 seconds of speech, tick the permission box, then start the clone.
The “Try MiniMax Speech 2.6 HD” button opens Deepnia’s voice tool with the model already selected.
Choose a MiniMax voice or one of your own, then paste up to 5,000 characters into “Text to speak”.
Choose the language and speed, check the cost shown, then click “Generate voice”.
Copy a prompt, open the tool and adapt it to your project.
Thank you for calling Linden Dental. We are open Monday to Friday, from 8:30 a.m. to 6:30 p.m. To book online, visit example.com/book. For emergencies, call (202) 555-0147. Please hold, and we will be with you shortly.
Hi Jordan! Your order #48213 left our warehouse on November 3 and should arrive between November 6 and 8. Track it at example.com/track or email us at help@example.com.
Hello, this is Lumière Salon with a reminder of your appointment on March 12 at 2:45 p.m. with Ines. To reschedule, call (202) 555-0147 at least 24 hours ahead. See you soon!
Welcome to Road Notes, episode 42. I’m Leah, and today we’re exploring night markets around the world. Show notes are at example.com/podcast. Let’s get started!
Attention, shoppers: the store will close in 15 minutes, at 9 p.m. Please make your way to the checkouts. We open again tomorrow at 8:30 a.m. Thank you for shopping with us.
Bienvenue au Hillside Hotel de Kigali. Le petit-déjeuner est servi en terrasse de 6 h 30 à 10 h. Pour le service en chambre, composez le 9. Une question avant votre arrivée ? Écrivez-nous à stay@example.com.
Deepnia’s other voice and music models, to pick the right one for your project.
| Model | Inputs | Settings | Best for |
|---|---|---|---|
| MiniMax Speech 2.6 HDThis model | Text, Authorized personal voice | Language, Speed | HD voice-overs, in your own voice if you like. |
| Open TTS | Text, Authorized personal voice | Language, Speed, Expression tags | Expression tags, plus a transcript or SRT captions with the audio. |
| ElevenLabs | Text | Language, Stability, Expression tags | ElevenLabs catalogue voices and three stability levels, in 10 languages or auto. |
| Gemini Flash TTS | Text | Expression tags | Expression tags with automatic language detection. |
| Deepnia Long-form | Text, PDF, Authorized personal voice | — | A whole document (PDF, Word, PowerPoint) read into a single MP3 file. |
Yes. French is in Deepnia’s language menu, and MiniMax documents French among the model’s languages. Listen to a short test to check how numbers and acronyms are read.
Yes. In the voice clone tool, record or upload 10–60 seconds of clear speech, confirm you have permission to use the voice, then pick it at the top of the MiniMax voice list.
23 on Deepnia, plus auto-detect, including English, French, Spanish, German, Arabic, Hindi, Japanese and Brazilian Portuguese.
5,000 characters per generation on Deepnia. For a longer script, split it into several generations.
Yes, from 0.70× to 1.20× in 0.01 steps, 1.00× by default.
Use punctuation to shape the rhythm (commas and full stops mark pauses), keep numbers, dates and web addresses in their usual form, pick the language of your text and adjust the speed: a little slower for information people must note, a little faster for an upbeat announcement.
An upbeat voice set around 1.05× gives an announcement some drive; your own cloned voice embodies your brand. Listen to a short test, then fine-tune the speed between 0.70× and 1.20×.
One WAV file (24 kHz, mono) per generation, ready to download.
The voice generator opens with MiniMax Speech 2.6 HD already selected. The exact cost shows before every creation; you sign up when you generate.
