SonaFactEvidence from audio
SonaFactEvidence from audio

From recording to transcript, AI minutes and subtitles

One file, as far as you want to take it: a transcript you can trust, minutes with receipts, a scored call review, subtitles translated into four more languages — or simply a clean DOCX.

Upload & processingTranscript editorAI documentsCall analyticsSubtitles & glossariesSearch & ask
Upload & processing

Drop a file and watch it work

Up to 20 recordings at once, audio or video — up to 30 minutes each on Free, 3 hours on Starter, 6 on Pro and 10 on Studio. Each one reports the stage it is in and how much is left — no spinner that tells you nothing.

✓The language is recognised for you across all hundred, or you name it yourself when you already know.
✓Speaker count: automatic, or an exact number if you know it.
✓An account glossary of your own names, passed to the engine as a hint.
✓Plain-language errors with a fix, not a code.
✓Large files can be made smaller in your browser before sending — from a video only the sound is kept, and the original stays on your device.
✓No account yet? Try one file of up to 5 minutes and read the transcript with speakers — it is deleted after 24 hours unless you sign up.
✓Automatic deletion on every plan: just the audio or everything, 1 to 365 days after the upload, with an email 3 days before.
✓Choose up to 20 files at once. Each is checked against your plan before it is sent, and they are transcribed in the order you chose. If the connection drops, you can resume a large file where it stopped.
✓Install SonaFact on your phone from the website. On Android it appears in the Share menu, so a voice message from WhatsApp or Telegram goes straight to SonaFact.
✓Record in the browser: press Record, see the time and the sound level, pause and stop. The sound reaches us in parts as you speak, so if the page closes or the connection drops, what was recorded is kept.
MP3 · WAV · M4A · MP4 · MKV · OGG · FLAC · WMA · OPUS · OGA · AMR · 3GP · WEBM · MOV · AAC · M4V · AIFF · CAF · AVI · WMV · MPEG · MTS · TS · VOB · MXF · FLVRetry and logs per file
Queue3 files
42:10product-review.mp4Transcribing · 64% · ~2 min leftProcessing
18:24interview-01.mp3Separating speakers · 88%Processing
07:55support-call.m4aChinese (94%) · 2 speakersDone
—old_tape.wmaAudio file not found — locate fileError
Stages: decoding → speakers → language → transcription → word alignment → glossary → utterances.
Transcript editor

Fix a word, rename a speaker everywhere

Playback that highlights the line being spoken, inline editing of the text, and one click to move a line to the right speaker.

✓Click a line to play from it; click the text to edit it.
✓Low-confidence words are underlined so you know where to look.
✓Find and replace across the whole transcript, with word timings kept.
✓Speaker panel with speaking time and share; move a line to another speaker or merge two into one.
✓Clean verbatim view: read the transcript without “um”, “uh” and stuttered repeats, while the transcript itself stays word for word.
✓Export to DOCX, PDF, Markdown, TXT, CSV, SRT or WebVTT, or as a file ready for NVivo, MAXQDA or ATLAS.ti.
✓Check a quote: paste a sentence and see where it was said — changed, added and missing words marked, with the uncertain words there. Text matching, no AI, on every plan.
✓Share a recording by link, with or without the audio, for 7 or 30 days or with no end date, and turn the link off at any time. People with a free account can comment on any line.
✓Highlight text in four colours with a note, and download only the highlights as DOCX or TXT. Turn a stretch of up to 10 minutes into a clip: share it by link, or download its audio from the Starter plan.
Search & replaceFollow playbackMerge speakers
product-review · utterances9 utterances · 142 words
00:00Anna WhitfieldGood morning, Ms Weber. Thanks for making time today.
00:10Speaker 2I do have a few questions about the schedule.
00:27Speaker 2We need to check invoicing in KSeF.
00:44Anna WhitfieldYeah.Low conf.
Short interjections like “yeah” or “ok” are the most common thing to correct.
AI documents

Minutes, action items, chapters — with receipts

Generate a document from one of six built-in kinds. Every item carries a clickable timestamp, so you can hear where it came from, and items that point to no moment in the recording are removed before you see it.

✓Meeting minutes, action items, summary, chapters, sales and support call reviews.
✓Verification line: “Removed 2 items with no matching moment in the recording.”
✓Edited the transcript? Generate the document again and it includes your corrections.
✓Output language independent of the recording language.
DOCX · PDF · Markdown · CopyEdit as markdownRegenerate
Action itemsMeeting minutes · up to date
01:01Anna WhitfieldSend the migration summary to all participantsToday
00:23James CarterConfirm the Munich data hand-over date18 Sep
00:52unassignedBook the follow-up meeting in two weeks—
🛡 Removed 2 items with no matching moment in the recording.
Call analytics

Scorecards for sales and support calls

Talk share, pace, longest monologue, interruptions and questions, plus a criterion-by-criterion scorecard with justification and the moment each one was met.

✓Built-in BANT and support standard scorecards, or your own criteria.
✓Roles per speaker with an AI suggestion you can override.
✓Objections with the response given, and next steps with owners.
✓Sentiment at the start, middle and end of the call.
Editable criteriaRegenerate after role changes
Sales call · BANTScore 72%
04:12BudgetCustomer named a range for the 2027 budgetMet
07:48AuthorityDecision maker mentioned but not confirmedPartial
11:02NeedTwo concrete problems with the current workflowMet
—TimelineNo date discussed for the decisionNot met
Roles changed — regenerate to update the score.
Subtitles & glossaries

Subtitles checked before you download

SRT and WebVTT with readability rules, up to four translated languages per recording, and glossaries that keep names and jargon spelled your way.

✓Up to 42 characters per line and two lines (less for Chinese and Japanese); timing aims for 17 characters per second, and anything still faster is counted, with its moment.
✓Translated sentences the model was unsure of, or that changed in the transcript, are marked.
✓After a transcript edit, Update re-translates only the changed sentences.
✓Your own glossary of names and jargon, kept per language or used for all of them.
✓Read the whole transcript in a translated language right in the editor, switching between the original and each translation.
SRTWebVTTGlossary hinted to the engine before it listens
Subtitles · Chinese (Simplified)Ready · 1 to check
00:12Anna我们先从迁移状态开始。—
00:19James第一阶段将在十月完成。—
00:27James我们需要检查 KSeF 中的发票。To check
00:41Anna我今天就发送摘要。Changed
Subtitles: 18 · too fast to read: 4
Search & ask

Find any word — or ask a question — across all your recordings

Search every transcript in your account at once, or ask in your own words. Answers are built only from your transcripts, and every statement links to the moment it came from.

✓Search ignores capitals and accents (“zolty” finds “żółty”) unless you tick Match case, and filters by date, language and speaker.
✓It searches the text with your corrections and your speaker names, and a click opens the recording at that moment.
✓Ask about one recording, the ones you choose or all of them — the answer is written in the language of your question.
✓Statements without a link to a moment are removed, and the answer says how many. If the answer is not in your recordings, you are told so, with the closest passages.
✓Search uses no AI jobs, on every plan. Each question uses 1 AI job from your daily allowance.
Search all recordingsAsk your recordingsQuestion history
Who sends the German text to the translator, and when?Passages used: 2 · Recordings: 1
00:19Speaker 2Speaker 2 — called Tom in the meeting — sends the German text to the translator today.—
00:00Speaker 1“Good morning, Tom.”—
Every statement links to a moment in your recordings — none removed.

Example: a 70-minute PTA meeting

The PTA secretary records the meeting on a phone placed on the table (with the consent of everyone present). The recording is 70 minutes long, so it doesn’t fit the Free plan, where the limit is 30 minutes per file. On the Starter plan ($9 a month, files up to 3 hours) it looks like this:

SonaFact separates the voices automatically. You rename a speaker once, e.g. to “Chair”, and the name changes throughout the transcript. If the engine split one person into two, one command merges them, and a single remark can be moved to the right person.

The meeting minutes link every point to its moment in the recording. When someone asks “who volunteered to run the bake sale?”, click the point and listen. It works the same for a journalist checking an interview quote or a law firm going back to one sentence from a call.

A name the engine hears differently, e.g. “Shivon” instead of “Siobhan”, goes into the glossary as a “heard as” variant. In later recordings it is corrected automatically.

Accuracy & speakers

“Speaker 2” becomes a name

Speakers are separated automatically, and the line being spoken stays highlighted while you listen. When the engine gets one wrong, you move it to the right person in one click — or merge two speakers it split by mistake.

The highlighted word follows playback, so you can read and listen at once.
Every transcript says what share of its words the engine was unsure about, and underlines them — so proofreading takes minutes, not a second listen.
Rename a speaker once and it changes across the whole transcript.
Your glossary teaches the engine your names and jargon before it listens, whichever language you record in.
Explore the editor →
product-review.mp49 utterances · 142 wordsEnglish
00:00Anna Whitfield
Good morning, Ms Weber. Thanks for making time today.
00:10Speaker 2
Good morning. I do have a few questions about the schedule.
00:23Speaker 2
The first phase is meant to close in October.
00:33Anna Whitfield
I understand the concern. All migrations will be tested by the end of next week.
Speakers
Anna Whitfield
31 s · 57% · 5 utterances
Speaker 2
Click the name to change it everywhere
23 s · 43% · 4 utterances
AI that shows its work

Every item links to the moment it was said

Minutes, action items, summaries, chapters and call reviews. Tap or hover an action item to see the sentence it came from. If the transcript doesn’t support an item, SonaFact removes it.

Meeting minutesUp to dateEnglish output
🛡Removed 2 items with no matching moment in the recording.
Decisions
▶ 00:33All migrations will be tested by the end of next week.
▶ 00:45Two external consultants join for the busiest weeks.
Action items
WhoWhatDueMoment
Anna WhitfieldSend the migration summary to all participantsToday▶ 01:01
James CarterConfirm the Munich data hand-over dateFri, 18 Sep▶ 00:23
unassignedBook the follow-up meeting in two weeks—▶ 00:52
DOCXPDFMarkdownCopy
Source transcript
00:00Anna Whitfield
Good morning, Ms Weber. Thanks for making time for the review today.
00:23Speaker 2
The first phase is meant to close in October, but I am not sure the Munich team will have the data ready in time.
00:52Speaker 2
Then I suggest we meet again in two weeks and go through the progress.
01:01Anna Whitfield
Agreed. I will put together a short summary and send it to everyone today.
Tap or hover an action item — the sentence it was taken from is highlighted in the transcript.
Meeting minutes

Participants, decisions, action items and a short summary.

Action items

Editable table: who, what, by when, and the moment it was agreed.

Chapters

Time-stamped chapter list, ready to copy into a YouTube description.

Your own scorecards

Edit the built-in criteria or write your own, reusable across calls.

Call analytics

Review sales and support calls, not just read them

Talk share, pace, longest monologue, interruptions and questions — plus a scorecard you can edit to match your own process. Every criterion carries the justification and the moment it was met.

Sales & support use case →
72%BANT score
Rep 58%Customer 42%
Pace 148 wpmLongest monologue 2:14Interruptions 3Questions 11
BudgetMetCustomer named a range for the 2027 budget.▶ 04:12
AuthorityPartialDecision maker mentioned but not confirmed.▶ 07:48
NeedMetTwo concrete problems with the current workflow.▶ 11:02
TimelineNot metNo date discussed for the decision.▶ —
Objections & responsesNext steps with ownersSentiment start · middle · end
Subtitles · Chinese (Simplified)Ready · 1 to check · 1 changed42 chars/line · 17 chars/s
TimeSpeakerOriginalTranslationStatus
▶ 00:12AnnaLet's start with the migration status.我们先从迁移状态开始。—
▶ 00:19JamesThe first phase closes in October.第一阶段将在十月完成。—
▶ 00:27JamesWe need to check invoicing in KSeF.我们需要检查 KSeF 中的发票。To check
▶ 00:41AnnaI'll send the summary today.我今天就发送摘要。Changed
SRTWebVTTSubtitles: 18 · too fast to read: 4
Subtitles & translation

Checked before you download

SRT and WebVTT with readability rules: up to 42 characters per line and two lines per subtitle. Timing aims for 17 characters per second — subtitles are stretched into pauses or merged, and any still too fast are counted, with their moments listed so you can jump to them. Translate into up to four more languages and read every translated sentence next to the original, with uncertain ones marked.

Correct the transcript later and Update re-translates only the sentences that changed.

EnglishPolishGermanFrenchSpanishChinese (Simplified)
Input

MP3 · WAV · M4A · MP4 · MKV · OGG · FLAC · WMA · OPUS · OGA · AMR · 3GP · WEBM · MOV · AAC · M4V · AIFF · CAF · AVI · WMV · MPEG · MTS · TS · VOB · MXF · FLV

Output

DOCX · PDF · Markdown · TXT · CSV · SRT · WebVTT · NVivo · MAXQDA · ATLAS.ti · copy to clipboard

Languages

100 languages, detected automatically · 32 with word-level timing

It tells you where it is unsure

Every automatic transcript has weak spots — a name said quickly, a word lost under a cough. Most tools hand you a clean-looking wall of text and let you find them yourself. SonaFact marks them.

Underlined where it matters

Words the engine was not confident about carry a dotted underline. You proofread those, not the whole hour.

A number for this recording

Each transcript states what share of its words were marked — usually a few percent on clear audio, more on a noisy phone call. It is measured on your file, not quoted from a brochure.

The same in every language

All one hundred languages go through the same engine and the same marking. A recording in Tamil gets its share of uncertain words stated just like one in English.

Try it on a recording you already have

The free plan is enough to judge accuracy, speaker separation and the editor.

Start transcribing for free
New in SonaFact

Autumn 2026: call recording, subtitles burned into video, a subtitle editor and more

28 new features — from recording a meeting without a bot to an AI assistant that reads your recordings. Each one backed by the recording, like the rest of SonaFact.

Everything that's new →