09-15-24, 04:16 PM
|
#17
|
|
Argentinian Skipper
Join Date: Jul 2009
Location: Somewhere between Buenos Aires and the Bungo Suido
Posts: 5,229
Downloads: 3280
Uploads: 0
|
fitzcarraldo has made a Best of SUBSIM nomination.
Nomination Category: Nomination basis: Great application for SH3
Forum: SHIII Mods Workshop
Nominated Member: shadow_wxh
Nominated Post: SH3VoiceCommand v1.0 Official Release
Quote:
====================== SH3VoiceCommand v1.0 Official Release ======================
After nearly six months development and much time preparation in advance. I am proudly announcing that SH3VoiceCommand v1.0 is officially released.
SH3VoiceCommand is a speech to command interface dedicated to support Silent Hunter game series. It uses state of the art AI speech recognition engine and offers a much greater complexity and functionality which surpass its' predecessor and peers.
It offers these following features:- 🎈 Lightweight: Only adds about 20MB RAM in the 32bit section.
- 📣Responsive: The command delay can be as low as 500MS with proper setup & optimization.
- 👶Independent: No dependency on any third-party software or MOD (except the OS and graphic card packages).
- 🎮Integrated: Completely merged with sh3.exe running on parallel thread which has no interference with users' keyboard & mouse inputs
- 🔠Multilingual: Supports 99 languages (current only configured for English and German, finetuned for German).
- 💡 Awareness: Aware of players' current coordinates heading speed and depth etc., and act accordingly.
- 🔢Numeracy: It can interpret any number from 0 to 1,999,999.99 in English and German
- 🎓Intelligence: Able to comprehend basic arithmetic symbols and operators like + - * / sin() cos() tan() pi and e in the configuration.
- 🤨Tolerable: Allows small deviation from standard command phrase to achieve higher catch rate
- 📎 Flexible: User can add, remove, change any voice command as they like in the config file
- 🤗Trainable: The language models can be finetuned to improve accuracy and adjust to personal accent
Disclaimer:- This MOD is under CC BY-NC-SA license which it under the following terms:
- 👨💻Attribution — You must give appropriate credit , provide a link to the license, and indicate if changes were made . You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use. You may publish a modified version of this MOD however it must bear a different name e.g. you shall not publish "SH#VoiceCommand v2.0". (note)
- ⛔NonCommercial — You may not use the material for commercial purposes . You may not list this material on any paid to download website.
- 👨🏻🤝👨🏻ShareAlike — If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original. Any application derived from or incorporating this material must also be free.
- (note) # stands for any number between 0~9 currently targeting to support 3, 4 and 5
Minimum System Requirements:- Operation System: Windows10 64bit
- RAM: 12GB
- Graphic Card: Nvidia Graphic Card with CUDA 6.1
- VRAM: 4GB
- Sound Card: Any sound card with microphone input
To install the MOD to your SH3 simply download the RARs in the download links in the bottom section of this post, read through user manual section one to seven and follow the instructions.
Some of the In-Game Screenshots
Backend Service and Logs
System Architecture
Known Issues:- Everyone has unique voice characteristics. It is likely each individual player needs to construct their own version config and language models. This may require a lot of time and dedication.
- Due to decimal rounding and ocean current flow the confirmed command from the crew and actual command from user may have ±1 difference.
- Depending on the user's graphic card performance and 3D workload the respond from transcriber can be slow. If this happens, switching on a live streaming application or a screen recording application can greatly improves the processing speed.
- Upon exiting the game you might encounter invalid memory access error, e.g. exception on 0x5D9104F9 (C0000005, EXCEPTION_ACCESS_VIOLATION), this is caused by SDL2.dll not being unloaded in the correct sequence. Simply ignore this warning because the game process has been terminated anyway.
- (Personally, I am not very satisfied with the workaround fix above but in current conditions it is the easiest way without changing sh3.exe itself. If you have any recommendations, please contact me through subsim messages, any help is appreciated)
- If game crashed while using SH3VoiceCommand the share memory might be leaked and can't be retrieved again. The transcriber will indicate "Share Memory NOT Ready" and exit after several attempts. You might need to restart windows in order to load SH3VoiceCommand again.
- Some of the command phrases may not be very accurate, have grammar mistakes or even missing. If you have any recommendations, please feel free to leave a comment under the publication page of SH3VoiceCommand on subsim. Also state your reference regarding where is your source of information.
References & Credits- https://openai.com/index/whisper/
Main web page of OpenAI Whisper
- https://github.com/ggerganov/whisper.cpp/
Open-source C++ version of Whisper
- https://cdn.openai.com/papers/whisper.pdf
Research paper of Whisper
- https://huggingface.co/blog/fine-tune-whisper
Detailed guide on how to fine tune Whisper by Sanchit Gandhi
- https://huggingface.co/datasets/mozilla-foundation/common_voice_11_0
Mozilla Common Voice 11.0 dataset
- https://huggingface.co/bofenghuang/whisper-medium-cv11-german
Bofenghuang's model after training with Mozilla Common Voice 11.0
- https://huggingface.co/datasets/shadow-wxh/VoiceCommandAudio
shadow_wxh's dataset on fine tuning SH#VoiceCommand
- https://huggingface.co/shadow-wxh/voicecommand-german-medium
shadow_wxh's model after fine tuning with VoiceCommandAudio
- https://maritime.org/doc/subphrase/
Some of the English command phrases are referenced from this document
- https://translate.google.com/?sl=en&tl=de&op=translate
Some of the German command phrases are translated using google translate
- https://www.subsim.com/radioroom/showthread.php?t=215433
The menu config file is derived from Ahnenerbe WideGui
- https://www.subsim.com/radioroom/showthread.php?t=132228&highlight=sh3speech
Some of the command phrases are inherited from SH3Speech
- Das Boot -Filmed by Wolfgang Petersen 1981
Some of German command phrases and taken from this film
- https://lostsouthpacificadventures.blogspot.com/2014/09/15000-miles-from-past-3.html
Photo of U-Boat speaking tubes
- https://www.flickr.com/photos/lensjoy/21528336803/
Speaking tubes onboard ORP Błyskawica taken by Marcin Bieszczanin. The voice pipe icon is edited from this.
- Photos of speaking tubes on board U-995 taken by Toyotagt86.
This is used as publication front page photo
- https://store.ubisoft.com/us/silent-hunter-3/5704fac788a7e32b078b4692.html?lang=en_US
Last but not least the original Silent Hunter III developed/published by Ubisoft January 2005
🙏Special thanks to Marcin Bieszczanin and Toyotagt86 for permission to use their excellent photographs on board ORP Błyskawica and U955🙏
OneDrive Download Links:
SH3VoiceCommand.part1.rar https://1drv.ms/u/s!AsHzobklG8RTjVom...JL8jI?e=qiswuQ
SH3VoiceCommand.part2.rar https://1drv.ms/u/s!AsHzobklG8RTjVwa...x8LC3?e=nFJgrZ
SH3VoiceCommand.part3.rar https://1drv.ms/u/s!AsHzobklG8RTjVvf...3FAiW?e=bBJYWc
SH3VoiceCommand.part4.rar https://1drv.ms/u/s!AsHzobklG8RTjVkK...IJYzu?e=Xdumcm
SH3VoiceCommandHotFix.rar https://1drv.ms/u/s!AsHzobklG8RTjV_b...3v-5r?e=USCX3M
FTP Download Links: (Limited Time & Connection Only, IP Address Subjected to Change)
SH3VoiceCommand.part1.rar ftp://Anonymous@123.191.118.202/SH3V...mand.part1.rar
SH3VoiceCommand.part2.rar ftp://Anonymous@123.191.118.202/SH3V...mand.part2.rar
SH3VoiceCommand.part3.rar ftp://Anonymous@123.191.118.202/SH3V...mand.part3.rar
SH3VoiceCommand.part4.rar ftp://Anonymous@123.191.118.202/SH3V...mand.part4.rar
SH3VoiceCommand.part5.rar ftp://Anonymous@123.191.118.202/SH3V...mand.part5.rar
|
|
|
|
Linear Mode
