Show notes need timestamps, the website needs a text version so the episode shows up in search, the newsletter needs two good quotes, and the video version needs captions. Doing all of that by hand for a fifty-minute episode takes an afternoon. The Voicepad bot in Telegram takes the episode file or a link to the published video and returns a transcript where every line has a speaker label and a timecode, with buttons for everything else on the list.
If the episode is on YouTube, send the link and tap Summary: the chapters at the end come as a timestamp and a short title, ready to paste into the description. Rename the titles to match your style; the timestamps themselves rarely need fixing. If you sent a file, there is no video to jump to, but every paragraph of the transcript has a timecode, so a chapter list is a matter of scrolling through and noting where the topic changes.
To find a quote, ask the Question button: "what did the guest say about hiring?" returns the answer, and the timecode next to the relevant passage leads to the exact wording in the transcript. Speaker labels are generic, Speaker 1 and Speaker 2, so before publishing replace them with names in one find-and-replace pass in the DOCX. When the host and the guest talk over each other or laugh at the same time, a line boundary can move by a word, which is worth a glance in the liveliest parts of the conversation.
Podcasts are recorded on decent microphones more often than not, and that is the best case for speech recognition. The bot runs on ElevenLabs Scribe v2: in our test on a Russian clip full of technical terms it made roughly half as many mistakes as Whisper-based bots, which turned "backend" into "backing". The language is detected automatically, English and Russian work best, and an episode where the guest switches between the two is fine. Still worth checking before publishing: guest names, company and book titles, numbers. A remote guest on a poor connection will produce more errors than a studio recording.
15 minutes every month are free, including the summary and one question per recording. That is enough to test the quality on a segment, not for a full episode. Packs: 60 minutes for 199 RUB, 300 minutes for 690 RUB (about six fifty-minute episodes), 1000 minutes for 1990 RUB. Minutes never expire, you pay for the length of the recording. Payment by any bank card in the browser, the /buy command in the bot.
Yes, lines are labelled Speaker 1 and Speaker 2 with a timecode on each. Rename the labels in the DOCX before publishing.
For an episode on YouTube, tap Summary: chapters with timestamps are at the end. For a file, use the timecodes in the transcript.
MP3, M4A, WAV, OGG, MP4, MOV and anything else ffmpeg can read, up to 2 GB.
Proportionally longer than a short clip, which is ready in about a minute.
15 minutes a month are free. A full episode needs a pack, the smallest is 60 minutes for 199 RUB.
Yes, the SRT button under the transcript returns a subtitle file with timestamps.
YouTube video to text with chapters, lecture notes from a recorded lecture or video, meeting recording to notes and action items, Telegram voice message to text. This guide in Russian: расшифровка подкаста.
Voicepad: voice, audio, video and YouTube to text in Telegram. 15 free minutes every month.