What it is
The Swiss army knife for video and audio โ it converts, cuts and analyses media files in practically any format.
How we use it
We check the real parameters of an audio or video stream (not what the description says) and process media in bots' background jobs.
Where it helps a business
- A bot accepts voice messages, video or music, and the file needs checking, compressing or converting.
- Telegram does not let heavy files through, while the customer needs the result in the chat.
- You need the real audio parameters, not what the description claims.
How we use it
- Checking and compressing before AI. In the voice transcription bot ffprobe checks the duration and whether there is sound, and FFmpeg recompresses files over 20 MB to mono MP3 at 32 kbps: enough for speech recognition, at a fraction of the size.
- Under Telegram's limit. In the video downloader bot FFmpeg merges tracks, produces MP3 with cover art, compresses and splits files into parts, and stitches TikTok photo slideshows into an mp4 with the post's music, normalising different aspect ratios to 1080ร1920.
- True quality. In ToshaMusic ffprobe reads the codec, sample rate, bit depth and bitrate of the downloaded file, and the name and tags are built from measurements only.
- Voice messages in chat. In Peregovorka the server transcodes a voice message to mp3 so iPhone, Android and desktop can all play it.
Common problems
- ffprobe misreports FLAC. For FLAC it often reports the container's 32 bits instead of the real 24, so the bit depth is refined separately.
- Compression kills quality. If compression would make quality unacceptable, the bot offers to split the file into parts instead of ruining it.
- Heavy work blocks the bot. FFmpeg runs in separate processes, and the number of concurrent downloads is capped.
When you do not need it
If files are only stored and forwarded as they are, there is nothing to process.