Make Your Voice Say Words You Never Recorded: IndexTTS 2.5 Voice Cloning on Your PC
Ultimate Text to Speech Generator clones any voice from a short clip and trains your own voice; IndexTTS 2.5 made 31 s of speech in 18 s on an RTX 5090.
Published Oct 02, 2026 · 1,749 views · Watch on YouTube
Get the app and files

Ultimate TTS: Voice Cloning & Training App With IndexTTS 2.5 & OmniVoice
Ultimate Text to Speech Generator clones any voice from a short clip and trains your own voice; IndexTTS 2.5 made 31 s of speech in 18 s on an RTX 5090.
Chapters
- 0:00 Make your own voice read new words
- 1:17 Choose reference cloning or a trained voice
- 1:51 Find the installer and Windows requirements
- 2:46 Move and extract the installer package
- 3:39 Install and open IndexTTS
- 4:45 Make and save your first voice clone
- 6:08 Build a clean reference from audio or video
- 7:08 Write longer narration and control the pauses
- 9:50 Save pronunciations for technical words
- 10:59 Give the same voice a different delivery
- 13:08 Adjust pace and fit a production slot
- 14:55 Change the useful quality controls
- 16:30 Export and reuse the finished speech
- 17:44 Turn subtitle cues into speech and video
- 19:27 Generate several scripts in one batch
- 20:43 Save and reopen your working setup
- 21:44 Start voice training from raw recordings
- 23:20 Create complete speech clips
- 24:56 Check the data and build the feature cache
- 26:25 Configure a new LoRA or DoRA voice
- 27:48 Save checkpoints and watch the training
- 29:43 Read the training result and automatic comparisons
- 30:55 Compare voices fairly and choose a checkpoint
- 32:39 Generate with the trained voice and its profile
- 34:10 Match the runtime to your GPU memory
- 35:28 Keep the app ready for the next project
- 36:20 Use the finished voice and choose your next project
About this video
Make your own voice speak words you never recorded. This complete IndexTTS 2.5 Windows tutorial starts with real generated speech, then takes you through a fresh installation, reference-based voice cloning, longer narration, expression controls, timed video, folder batches and training a voice from original recordings. Follow the actual local application from a short sample and script to saved audio and a trained voice setup.
Links:
Download IndexTTS 2.5 and read the full feature guide: [ https://www.patreon.com/SECourses/posts/indextts-2-5-and-139297407 ]
Windows Requirements tutorial: [ https://youtu.be/DrhUHnYfwC0 ]
Updated Windows Requirements instructions: [ https://www.patreon.com/posts/windows-AI-requirements-tutorial-111553210 ]
You will learn how to find the current ZIP, locate it in Chrome, move it to a permanent folder and extract the whole package. Complete the linked Windows requirements first. Choose a path using English characters with no spaces, then run Windows_Install_or_Update.bat and Windows_Start_App.bat normally. Double-click the BAT files or select one and press Enter; do not run them as administrator. Keep the app command window open while you use its local browser interface. Load a clean voice reference, enter a short script, generate speech, play the complete result and find the output files. The reference can come from audio or video; use the range controls to choose the useful speech. Smart sentences groups longer scripts, while explicit pause tags and pronunciation dictionary entries help shape the result.
Main topics include speaker reference versus emotion reference, emotion vectors and text-driven expression, speaking rate and timing modes, candidate generation, saved outputs and reusable presets. Create a complete video from an SRT file and background image, with each caption tied to its intended time. Generate separate named audio files from a folder of scripts. For training, start with raw recordings and matching captions, prepare clips, audit the dataset and cache features. Keep entire source recordings separate for training, validation and the final test. Configure DoRA training, follow its loss and speech checks, inspect the saved checkpoint recommendation and compare Base with selected checkpoints using the same text, reference, strength and seed. The voice decoder has its own validation decision. A saved voice profile can carry reference, pace and decoding choices back into Voice Generation. Generate new words with the trained voice, inspect its line-length and pause guidance, and save the speaking rate for that adapter. Models and Performance covers GPU memory tiers, standard and INT8 ConvRot weights, block swapping, auxiliary model residency and the isolated memory benchmark. Model-file verification, unloading, Help, Changelog and the normal update route finish the Windows workflow.
Use the chapters to follow the installation in order or jump to voice cloning, expression, caption video, batches, training, checkpoint comparison and GPU memory controls.
Chapters:
0:00 Make your own voice read new words
1:17 Choose reference cloning or a trained voice
1:51 Find the installer and Windows requirements
2:46 Move and extract the installer package
3:39 Install and open IndexTTS
4:45 Make and save your first voice clone
6:08 Build a clean reference from audio or video
7:08 Write longer narration and control the pauses
9:50 Save pronunciations for technical words
10:59 Give the same voice a different delivery
13:08 Adjust pace and fit a production slot
14:55 Change the useful quality controls
16:30 Export and reuse the finished speech
17:44 Turn subtitle cues into speech and video
19:27 Generate several scripts in one batch
20:43 Save and reopen your working setup
21:44 Start voice training from raw recordings
23:20 Create complete speech clips
24:56 Check the data and build the feature cache
26:25 Configure a new LoRA or DoRA voice
27:48 Save checkpoints and watch the training
29:43 Read the training result and automatic comparisons
30:55 Compare voices fairly and choose a checkpoint
32:39 Generate with the trained voice and its profile
34:10 Match the runtime to your GPU memory
35:28 Keep the app ready for the next project
36:20 Use the finished voice and choose your next project
The tutorial demonstrates v6.22, with automatic model downloads, a prepared environment and prebuilt acceleration libraries. Read the training recommendations, listen to the complete checkpoint comparisons, and choose the voice setup that works for your project.
The package also supports RunPod, SimplePod and Massed Compute. This video teaches the Windows route; the linked post explains the other supported platforms and current downloads. The app link is also in the pinned comment. For installation questions, feedback and feature requests, use the post and video comments.