Q1:How many languages are supported?
A1:93 countries in total. English, Chinese, Korean, Japanese, German, French, Spanish, Portuguese, Dutch, Italian, Thai, Vietnamese, Indonesian, Hindi, Telugu, Malayalam, Bengali, Arabic, Urdu, Russian, Ukrainian, Hebrew and many others.
Q2:What is the translation accuracy?
A2: Sakura can use OpenAI's whisper-1 and Chat-GPT on the CPU as well as Faster-whisper on the GPU, so lightweight and highly accurate translations can be expected. However, not all languages can be translated with the same accuracy. For example, the following comparison table is linked for reference.
https://github.com/openai/whisper/discussions/2363
Q3:What is the real-time performance?
A3:Unfortunately, perfect simultaneous interpretation is difficult due to the nature of language grammar, but the PVT series uses 3 seconds as a standard, and by adding functions such as silence detection, translations are made in units of one phrase or one sentence as much as possible. The average delay is 1 or 2 sentences. The translation speed is about the level of “a little slow? Please be aware that there may be further delays depending on the network environment. In fact, we sometimes see cases where the server on the OpenAI side seems to be down. This is because the server side is responding on a best-effort basis. If you think that the server is frozen, you can kill the application and switch to Faster-whisper (transcript locally) to deal with the problem. Faster-whisper recommends GPU/CUDA environment.
Q4:Will the conversation be leaked to outside parties?
A4:Whisper-1 and Faster-whisper handle voice data differently. Whisper-1 sends voice data to OpenAI's server, while Faster-whisper processes it inside the PC, so it is not leaked on the network. However, the Transcripted text data will be sent to Google and OpenAI to use a separate translation engine, and will go through the network. Please avoid conversations about sensitive information. Also, we cannot guarantee opt-out.
Q5:How long can I use the trial version?
A5:15 days after installation and first execution.
Q6:How long can I use the commercial version?
A6:Basically, it is valid until the end of July 2030. The reason for this is that development in this field is very fast and performance and specifications change significantly, which may affect compatibility. We have taken the liberty of setting an expiration date of approximately 5 years after release. The trial version will be available until the end of July 2030 as well.
Q7:Where can I obtain the GPU environment?
A7:The PVT series supports CUDA; please register as a user separately from the NVIDIA website and install CUDA. CUDA Toolkit 12.x and cuDNN 8.x are recommended.
However, a full CUDA environment is not required; the PVT series will automatically install and run the appropriate CUDA libraries if NVIDIA GPU drivers are installed at a minimum. (Existing CUDA environments will not be affected, but older drivers may cause some instability. It is recommended to update to the latest drivers.)
How to check the driver:
Open a command prompt and execute “nvidia-smi”. If the driver Version is displayed, it is OK.
Example: C:\Users\<user name>\nvidia-smi
Reference: CUDA Toolkit Archive | NVIDIA Developer
Q8: What is Re-translation?
A8: Starting with the 2nd release v20250322, if the conversation proceeds without interruption for a certain period of time (default about 6 seconds), a forced break is established and a temporary translation is started. Then, at the next translation, the previous script is retained and consolidated, and the entire sentence is retranslated. This should be done as one turn, with up to two retries (excluding the first time). This function emphasizes translation accuracy rather than real-time performance. Note that the result of temporary translation during retry changes the color of the text to brown. If you do not need to retry translation, please edit the following envrionment.json file directly before launching the app. Note that if you edit the json file while the app is running, the current settings will be overwritten when the app exits and will not be reflected. Please be aware of this.
Ex：C:\Users\<User_Name>\AppData\Local\omicronware\PVTsakura\user_env\environment.json
"translator": {
"default_api": "gpt-4o-mini",
・・・・・・・・・・・・
"force_retranslation": "on" → "off"
},