331–340 of 383 posts
I hope this is the future. Offline, small ML models, running inference on ubiquitous, inexpensive hardware. Models that are easy to integrate into other things, into devices and apps, and even to drive from other models maybe.
Hmm. A pay once (or not at all) model that can run on anything? Or a subscription model that locks you in, and requires hardware that only the richest megacorps can afford? I wonder which one will win out.
Earlier quoted context omitted.
You're getting a lot of comments along the lines of "Why don't you just ____," which only shows how Stockholmed the entire Python community is. With no other language are you expected to maintain several entirely different versions of the language, each of which is a relatively large installation. Can you imagine if we all had five different llvms or gccs just to compile five different modern C projects? I'm going to…
I agree with your point, but > if we all had five different llvms or gccs Oof, those are poor examples. Most compilers using LLVM other than clang do ship with their own LLVM patches, and cross-compiling with GCC does require installing a toolchain for each target.
Earlier quoted context omitted.
Speech speed is always a tunable parameter and not something intrinsic to the model. The comparison to make is expressiveness and correct intonation for long sentences vs something like espeak. It actually sounds amazing for the size. The closest thing is probably KokoroTTS at 82M params and ~300MB.
I think he meant overacting typical for English dubs.
I ran some quick benchmarks. Ubuntu 24, Razer Blade 16, Intel Core i9-14900HX Performance Results: Initial Latency: ~315ms for short text Audio Generation Speed (seconds of audio per second of processing): - Short text (12 chars): 3.35x realtime - Medium text (100 chars): 5.34x realtime - Long text (225 chars): 5.46x realtime - Very Long text (306 chars): 5.50x realtime Findings: - Model loads in ~710ms - Generates a…
Hmm the quality is not so impressive. I'm looking for a really naturally sounding model. Not very happy with piper/kokoro, XTTS was a bit complex to set up. For STT whisper is really amazing. But I miss a good TTS. And I don't mind throwing GPU power at it. But anyway. this isn't it either, this sounds worse than kokoro.
Hmm the quality is not so impressive. I'm looking for a really naturally sounding model. Not very happy with piper/kokoro, XTTS was a bit complex to set up. For STT whisper is really amazing. But I miss a good TTS. And I don't mind throwing GPU power at it. But anyway. this isn't it either, this sounds worse than kokoro.
Try https://github.com/Picovoice/orca
Hmm the quality is not so impressive. I'm looking for a really naturally sounding model. Not very happy with piper/kokoro, XTTS was a bit complex to set up. For STT whisper is really amazing. But I miss a good TTS. And I don't mind throwing GPU power at it. But anyway. this isn't it either, this sounds worse than kokoro.
You should give try to https://pinokio.co/
Hmm the quality is not so impressive. I'm looking for a really naturally sounding model. Not very happy with piper/kokoro, XTTS was a bit complex to set up. For STT whisper is really amazing. But I miss a good TTS. And I don't mind throwing GPU power at it. But anyway. this isn't it either, this sounds worse than kokoro.
The best open one I've found so far is Dia - https://github.com/nari-labs/dia - it has some limitations, but i think it's really impressive and I can run it on my laptop.
But it might be worth setting it up and seeing if it improves over time.
Hmm the quality is not so impressive. I'm looking for a really naturally sounding model. Not very happy with piper/kokoro, XTTS was a bit complex to set up. For STT whisper is really amazing. But I miss a good TTS. And I don't mind throwing GPU power at it. But anyway. this isn't it either, this sounds worse than kokoro.
Imho chatterbox is the current open weight SOTA model in terms of quality: https://huggingface.co/ResembleAI/chatterbox