- This topic has 5 replies, 1 voice, and was last updated 47 minutes ago by
.
-
Posts
-
Have something to say about NexSub? Say it here!
Have suggestions, comments, or need help? Post it here! If you know of better software than NexSub, post it here! If you know of issues with NexSub, post it here! Share your knowledge with all of us. :-)
MANY modern computers do not have a stereo mixer input, it’s not just disabled or hidden it does not exist in hardware. It was pushed from Vista onwards as an effort to protect digital media from being recorded practically losslessly… they did not realise that WASPI introduced a pure software way to perform the same function and sadly THIS product is based upon windows XP recording API so cannot work on modern systems without a hardware and driver support for Sound Mixer recording input. I was going to get this but I just noticed the download size for just the installer 2.9GBytes that WILL expand significantly on installation… whilst I do have maybe a TByte spare over a few physical drives this will by default write to the system drive where all the browser caches are … too much of a gamble for me! I also notice on their product page they write:
“Minimum
Intel / AMD / NVIDIA
RTX 3050 8 GB or equivalent”While here they say it must have a NVIDIA GPU with no tolerance for Intel or AMD equivalence and it looks like that although I have a dedicated NVIDIA GPU it’s only 4GB dedicated VRAM so looks like I’d probably be out of luck anyway! They don’t even host the download themselves but send us to MajorGeeks.com to get it! And on Majorgeeks.com product page it says the price is just $9.00 not $19.00 someone’s not paying attention… Probably just as well this is likely written by an AI coding agent
I repented and figured out a way to get this without causing multi-gigabytes of unwanted writes to my poor system drive… Well their system requirements are over stated… It works in Nvidia Nvidia GTX1650 with only 4GB of dedicated VRAM AND it only uses around 2GB of video RAM! I used Ant download to download it to a non-system drive and installed it into another non-system drive so as to not shorten the lifetime of my system. It does NOT remember your chosen record input.. and does NOT use the one that is set to windows default record input I had to manually select the Stereo Mix input EVERY time I started the program. You are required to play the video in an external video player and it renders the subtitle as just a single line of text over the top of all the windows including a full screen playback, because the processing is in real-time and not pre-processing the sound track the subtitles are ALWAYS lagging behind the videos sound track it is unavoidable and unhelpful and confusing if you use subtitles but are not totally deaf . Also there is NO WORD-WRAP for long sentences… the subtitle can fall off the edge of the screen… it doesn’t care! There is no centering of the text rendered it has a fix non adjustable left margin that is practically a quarter of the screen width.
It needs to read the videos soundtrack and pre-process the audio and generate real SRT files that are then combined with the video/audio file so the subtitles are synchronised this is an idea that was not thought out before they told the AI coding agent to generate it! It does have an export to SRT but I’d prefer it do the SRT generation directly from the media file and not by recording the system sound… don’t forget this makes it impossible to process at greater than real-time playback rate and makes it likely the SRT timing values will be inaccurate and will require manual editing to fix, especially fixing the lack of word-wraping that will have to be inserted manually. The translation log keeps resetting to the top of the log instead of auto-scrolling to keep track of current output. And the GPU usage is pulsing not consistent. Also it DOES NOT TRANSLATE the spoken text into another language as the description and user interface implies it remains in the spoken language just makes a mess of it because it’s using the term translate incorrectly… they mean “transcribe” and not “translate” the two are very different English words with VERY different meanings… Even AI TRANSLATORS should know the difference. You need to chose the spoken language to get a reasonable TRANSCRIPTION it DOES NOT TRANSLATE languages from one to another! I’ll probably uninstall this without registering as the 7 day trial is not crippled as far as I can see it’s just badly implemented.
And because the program is not aware of the video player program or the input video file name the exported SRT file default file name is less than useless and you need to make it match the playing videos file name to have a chance of using it especially if you have multiple similar looking videos…
It would be better if it included it’s OWN legal video player that the video had to be played in so it could obtain the details of the playing video to name the SRT file automatically CORRECTLY.
It also seems quite happy to swear inappropriately so obviously has been trained on material with considerable amount of swearing so it’ll guess F’n this and that when it’s not sure of the real words. This needs an alternate CLEAN set of training data that WON’T SWEAR EVER and can censor spoken swear words rather than transcribe them uncensored. It cannot be trusted to not swear out of the blue when no swear words are in the source material. So unsuitable for uncensored subtitling for people that want clean subtitles.
The program should be using WASAPI to sample the audio from the video playing process and not just the speaker output as other programs could make noises that would interrupt the played back audio. It should not be using discrete ancient windows 95/XP directsound input devices API.
I note it is possible to grab hold of the always on top subtitle rendering box and move it over the the far left to reduce the number of times the subtitles overflow off the right hand edge of the screen, it does NOT remember the custom location of the subtitle box so you will likely need to move it on every restart. There is no in program way to clear the transcription log area so if you want to transcribe two or more videos and export out SRT files for each of them, not forgetting the naming problem… you need to fully restart the program and output panel and recording device every time. Other similar programs also use whisper speech recognition model and are similarly inaccurate. It also repeats some words at the end of the previous line… onto the beginning of the next line. Maybe AI models are just incapable of accurately transcribing just like they cannot be guaranteed to only give correct answers if they are based on LLMs as they are all based heavily on flawed statistical probabilities and are not actual knowledge based expert systems.
oh all steelsoft AI products vomit out a few gigabytes of TEMP files onto the system drive when processing and leaves them there when the program exits containing many subfolders with python files and compiled python and copies of the AI model inferences rather than keeping them where they are chosen to be installed. They are not built to be easy on the systems they’re installed on. Note you can bypass their 7 day trial limitation by using nirsofts RunasDate.exe the 64bit version, why am I telling you this because they blatantly steal GPLv3 FFMPEG binaries on some of their products so if they cannot comply with the copyright licenses of products they plagiarise they shouldn’t expect anyone to honor their copyright restrictions on their illegal closed source combined works.
[@TK] Thanks, TK, for such a clear, in-depth review.
[@TK] Thanks a lot for your great review and discussion of very, very important deficits of the software! You saved me from downloading and using this software – as for me – useless.
[@Mr.Dave]
I’m surprised Mr. Boss doesn’t prequalify the
software he promotes.2.9 GB !!!
