A little guidance.
Back to your work.

1. Create an account or sign in, complete your profile and activate your subscription. After payment is confirmed, your account page provides the Windows download.
2. Run the downloaded WhoSaid installer. Read and accept the license agreement, then finish installation and open WhoSaid.
3. In the app, open Settings → WhoSaid account → Connect account. The app displays an App connection code and opens the account website.
4. Sign in with the same account used for your subscription. Enter the code and approve the connection. Return to WhoSaid; use Check subscription if access has not refreshed.

You do not need your own Daily, Azure or Deepgram account, or a provider API key. If the connection code expires, choose Connect account again.

The desktop app is for Windows. Clients can sign documents and join calls from a supported browser on a phone or computer without installing WhoSaid.

1. Connect your account while you have internet access.
2. For local transcription, select Run on CPU (or an available NVIDIA GPU) and choose your model in Transcribe, then open Settings → Local transcription tools → Set up speech tools. Keep at least 8 GB of free space available during setup. Downloads can take a while; later jobs reuse them.
3. In Record and Transcribe, choose your audio sources and run Test 10 Seconds. Listen to the result.
4. Check your recording and transcript folders so you know where the files will go.

WhoSaid prepares CPU speech tools automatically when needed. You do not need to install Python or obtain a Hugging Face account. Compatible NVIDIA acceleration may require additional setup; see the GPU guide. Online transcription does not need local speech-model downloads.

Make sure you have the required permission to record and, if applicable, upload the conversation.

• Record and Transcribe: capture your microphone and computer audio, then transcribe the recording.
• Transcribe: turn an existing audio or video file into a transcript.
• Video / GIF: trim a video, add arrows or circles, and export MP4 or GIF.
• PDF / Signatures: organize and mark up a PDF, add Bates numbers, or send it for a drawn signature.
• Calls: invite one person to a browser audio call with captions and translated text.
• Settings: manage your account connection, speech tools, appearance and app updates.

Help and Contact Us remain at the upper right of the app.

1. Open Record and Transcribe.
2. Under Audio sources, set System audio to the speakers or headphones playing the remote participants. Set Microphone to the input carrying your own voice.
3. Select Test 10 Seconds. Speak and play the other audio while watching the two meters. The test stops automatically. Select Play last recording and listen for both sides.
4. Check Save folder and File name prefix, then select Start Recording. Select Stop Recording when finished.

Use Refresh devices after reconnecting equipment. Run another short test whenever you switch headphones, speakers or microphones.

Computer audio and microphone audio are mixed into one WAV recording. A Zoom connection may contain many voices; WhoSaid estimates speakers from their voices, not from an assumption that each audio source is one person.

Select your processing option, then choose Transcribe Last Recording. Or select Transcribe automatically after stop before recording. The online option still asks you to approve the upload and estimated charge.

This feature transcribes after recording; it does not display a live transcript while recording. Live translated captions are a separate feature in Calls.

Transcribe Last Recording refers to the current app session. After restarting WhoSaid, use Transcribe → Open recording… to select your saved WAV.

Choose Open recordings folder. The default is Documents → WhoSaid → Recordings. WhoSaid saves WAV files and adds a date, time and unique suffix to the chosen prefix.

If the other participants are silent, check that System audio matches the output device actually playing them. If your voice is missing, check the selected Microphone, its mute switch and Windows microphone permission. Reconnect the device, choose Refresh devices, and listen to another ten-second test.

You can record only your microphone or only system audio by choosing the corresponding “None” option for the other source. At least one source must remain selected.

1. Open Transcribe → Open recording… and choose your audio or video, such as WAV, MP3, M4A or MP4.
2. Check Save transcripts. The default is Documents → Transcripts. Use Choose folder… to select another folder for this session.
3. Choose CPU, NVIDIA GPU or online processing. For local English transcription, leave Language at English and review the model and speaker settings.
4. Select Start transcription. Leave Open transcript in WhoSaid when done selected to open the result in the reader.

The source recording stays unchanged. TXT is the readable transcript; SRT and VTT are subtitles; JSON and TSV contain additional timing and transcript information.

Comparing runs? Choose a separate output folder for each local run. Reusing the same source filename and output folder replaces the raw transcript exports with that name. Save your edited transcript as a separate copy.

CPU transcription and speaker identification can take as long as the recording or longer. A two-hour recording may take two hours or more. Speed depends on the computer, audio and model.

You can switch between jobs. Choosing CPU does not permanently disable GPU processing. Local processing has no online transcription usage fee, but your WhoSaid subscription and periodic account checks still apply.

WhoSaid enables this option after verifying compatible NVIDIA hardware and acceleration tools. Select Check graphics card and read the result. An Intel or AMD graphics processor does not satisfy this NVIDIA option.

If speech tools are missing, choose Settings → Set up speech tools. Automatic setup prepares CPU tools; it does not install a complete NVIDIA environment on every computer. If you have an NVIDIA card but the result says your speech tools support only CPU processing, contact us for help.

You can keep using CPU or online processing. You do not need to uninstall and reinstall WhoSaid merely to change processors.

Small is the default. Tiny and base are lighter choices; medium is the largest available and usually requires more time and memory. These choices affect recognition of the words. All four use the same Community-1 speaker-identification model.

Leave Identify speakers (Community-1) selected to label speakers. If you know there are exactly four speakers, put 4 in both Minimum speakers and Maximum speakers. Otherwise, use a plausible range; the defaults are 2 and 8.

Similar voices, short replies, overlapping speech and poor audio can cause mistakes. Speaker labels are estimates, not verified identities. Review the transcript and use the reader’s speaker-name tools to correct it. A larger speech model does not select a different diarizer.

These settings apply to local processing. The online option uses its own English transcription and speaker identification.

1. Select Online: Deepgram ($1/audio hour), then start transcription.
2. WhoSaid prepares an audio-only copy locally. Review the recording duration, estimated charge and upload disclosure.
3. Approve the dialog to upload. Keep WhoSaid open while it waits for and saves the result.

The service currently supports English recordings up to four hours long. The rate is $1 USD per recording hour, so 30 minutes costs $0.50. Processing time does not determine the price. Successful usage is recorded for a separate invoice; starting a job does not immediately charge your card. Automatic usage invoicing is not active yet.

No audio is uploaded before you approve. WhoSaid removes uploaded audio after processing and keeps encrypted results available for pickup for up to seven days. Downloaded transcripts remain in your chosen local folder. Provider retention policies still apply.

Stop transcription stops waiting and attempts to cancel work that has not begun. Already-started processing can still finish and be billable. If interrupted, retry the same file in the same output folder to retrieve the existing job. Keep its .whosaid-online receipt folder. WhoSaid asks separately before starting a new paid attempt.

The reader opens automatically after transcription if Open transcript in WhoSaid when done is selected. To reopen a transcript, choose Transcribe → Open transcript… and select its TXT file.

1. Edit the words directly in the text area.
2. Enter Case name, Case number, and Caption / court / parties in the sidebar. These appear above the saved or printed transcript.
3. Choose Save edited copy. The suggested name ends in - edited.txt to preserve the machine transcript. Later saves use Save changes; use Save as… for another copy.

Keep the TXT and its adjacent .txt.whosaid.json file together when moving or backing them up. The companion preserves case details, original speaker IDs and name mappings. The TXT can still be read without it.

1. Under Speaker names, review up to four excerpts for each speaker ID.
2. Type the person’s Actual name. If two IDs belong to the same person, give both the same name.
3. Select Apply names / refresh excerpts, review the result and save your edited copy.

The reader keeps the original ID visible alongside the name while you review; the exported copy uses the chosen name. Naming an ID does not automatically fix a passage assigned to the wrong speaker. Correct those mistakes in the transcript text.

Reader edits affect your edited text, not the original JSON or subtitle exports.

Select Print… to choose a Windows printer, or Save PDF… to create a PDF directly. Microsoft Print to PDF may also be available in the Windows printer dialog.

Line numbers run continuously and count actual line breaks. Wrapping a long line on a narrow screen does not create an extra numbered line. Reflow long lines splits long text into shorter review lines; Ctrl+Z can undo that change. PDF line numbers continue across pages.

Saving a PDF does not also save the editable TXT. Save the text as well to retain your work for later editing. Use Open Transcript Folder to find the saved files.

Shortcuts: Ctrl+S saves, Ctrl+Shift+S saves another copy, and Ctrl+P prints. These are machine-generated working transcripts and require review.

1. Open PDF / Signatures → Open PDF..., then use the Edit PDF tab.
2. Select a page and use Move up, Move down, Rotate left, Rotate right, Duplicate or Delete.
3. Choose Open large PDF editor. Use the page controls and Zoom to see the area clearly.
4. Under Markup, choose Pen, Text, Rectangle or Arrow, then choose a color and size. For text, enter Text to place and click its position on the page. Draw arrows from their tail toward the point you want to indicate.
5. Choose Save copy... and a new filename. Use Open saved PDF or Open folder to see the result.

Done closes the large editor but does not save a PDF. Save your copy before closing WhoSaid. Undo last change reverses recent edits.

The editor adds text and markup; it does not rewrite existing PDF paragraphs like a word processor. Password-protected PDFs are unsupported. Keep the original of any digitally signed PDF, since editing can invalidate that signature.

1. In the large editor, choose Rectangle and check Fill rectangle.
2. Choose a color, including white if wanted. Drag a filled rectangle over the entire area to remove.
3. Review all covered areas, choose Save copy..., then open and inspect the saved PDF.

A filled rectangle causes WhoSaid to rebuild the entire saved PDF as image-only pages, removing the covered source content from that copy. This also removes selectable text, forms, links, attachments, metadata and existing digital signatures. The original remains unchanged.

White pen strokes and unfilled rectangles are only markup. Use the filled-rectangle workflow when content must be removed from the exported copy.

Open Bates numbers in the large PDF editor. Set Start on page, First number, an optional Prefix, and Number digits. Choose Apply Bates numbers, then save a copy.

For an unnumbered cover sheet followed by page 1, set Start on page to 2 and First number to 1. For example, prefix EXH- with six digits produces EXH-000001.

Remove Bates numbers removes numbering added in the current editing session before saving.

Under Finish and send, choose Create email draft.... Save pending PDF edits when prompted. Enter To, Subject and your message, then choose Save and open draft.

WhoSaid creates an .eml draft with the saved PDF attached. Review and send it in a compatible email app. Some webmail programs do not open these drafts directly; you can always attach the saved PDF yourself.

This button prepares an email; it does not send it automatically. Sharing documents through a storage link is not available in this version.

1. Open the finalized PDF in PDF / Signatures. Save pending edits, then open Sign & Send.
2. Enter Client email and Your email for signed copies. Choose Save email to remember your delivery address.
3. Select Send signature request. The client receives a private link by email. Copy the eight-digit access code from the one-time window and share it separately from that email.
4. The client opens the link, enters the code, reviews the PDF and draws with a finger, mouse or stylus. They choose Place drawing on this page, position the signature, acknowledge the notice and select Send signed document.

The completed PDF is emailed to the attorney and client. The client’s completion page also offers Download signed PDF while the copy remains available. The client does not need an app installation or a paid account.

This supports one signer and one drawn placement. It does not verify identity, notarize, provide a signature certificate or produce a signing audit trail. The signing copy becomes static image pages.

Completed requests leave Active requests, which lists unfinished requests. In Sign & Send, choose Check for signed PDFs, then Open signed PDFs folder.

WhoSaid checks approximately every 30 seconds while open. The default folder is Documents → WhoSaid → Signed PDFs, using your actual Windows Documents folder, including a OneDrive location if Windows redirects it there.

If the copy is missing, keep the app connected, check disk space and folder access, and retry. Check spam and the two delivery email addresses. Email acceptance by a provider does not guarantee that the message appears immediately in an inbox.

Completed PDFs wait encrypted online for up to 30 days for pickup and can be removed earlier after successful desktop pickup and email acceptance. Keep and back up your downloaded copy; this is temporary delivery, not an online document archive.

Unfinished signing requests expire after 48 hours. The eight-digit code appears only once. If it is lost, cancel the unfinished request with Cancel selected request and create a new one. A completed signature cannot be cancelled through that button.

Repeated wrong codes can temporarily lock the signing page. Check the intended code before trying again; the usual lock period is 15 minutes.

Signing uploads are limited to 25 MiB and 200 pages, with additional processing limits for unusually large pages or complex files. Password-protected PDFs are unsupported. Split a very large document or contact us if it cannot be prepared.

1. Open Video / GIF → Open video.... You can also paste a local file path and choose Load path.
2. Move the preview slider. Use the Start and End Set buttons, or type seconds, minutes:seconds or hours:minutes:seconds.
3. Choose Export MP4... or Export GIF... and save to a new filename. Wait for completion, then choose Open export or Open folder.

MP4 includes available source audio. GIF has no sound. Smaller GIF dimensions and lower frame rates usually produce smaller files. Defaults are 720 pixels wide and 12 frames per second.

The preview lets you inspect frames; it is not a full playback timeline with sound. Your source video stays unchanged.

Set the start and end times for the highlight. Under Highlights, choose Arrow or Circle, a color and Width (px), then drag on the preview.

Each highlight uses the time range selected when you draw it. Use Delete, Undo or Clear to revise the marks. They do not automatically follow moving objects. Check the overall clip start and end before exporting.

If an export fails, open Export log. Finish an active recording or transcription first if WhoSaid reports another operation is busy.

1. Open Calls. Enter Client email, choose You speak and Your client speaks, and select two different languages.
2. Set Call length · minutes within the service limit, then choose Create invitation & open call. The client receives an invitation; your private attorney page opens in a browser.
3. Each person reviews the transcription and translation notice, selects the consent checkbox and chooses Agree and continue. Use Test microphone and allow browser microphone access.
4. The attorney selects Admit client. After consent, admission and microphone checks are complete, each person chooses Join audio call.
5. Speak in short phrases and pause for captions and translation. Use Type a clarification or Please repeat that when needed.

Both sides hear the original voices and see original text plus translations. Translations follow completed phrases; WhoSaid does not speak a translated voice.

Share the client invitation, and keep the attorney link private. Use End for everyone to finish the conversation for both people. Leave call allows rejoining while the invitation remains active.

WhoSaid 2.4.0 offers 63 language choices, including regional and script variants, using the supported Nova-3 speech and Azure translation combination. Type in the language selector to find a language. The in-app list is the reference for available choices.

The client’s invitation and page instructions use the selected client language. Display changes the page’s interface language, not the participants’ configured speaking languages.

The client needs internet, a microphone, speakers or headphones, and a current Chrome, Edge, Safari or Firefox browser on a phone or computer. They do not need WhoSaid, a paid account, an NVIDIA card or speech-model downloads.

Calls use online Daily/Deepgram speech processing and Azure translation. Rooms support two people and do not provide a telephone dial-in number. Translated calls are separate from English file transcription.

• No microphone: allow microphone access in the browser’s site permissions, select the correct input and run Test microphone again.
• Echo or poor captions: use headphones, move closer to the microphone, reduce background noise and avoid speaking over each other.
• Missing or wrong translation: confirm both speaking languages, use shorter phrases and type a clarification for names or critical details.
• Disconnected: pause speaking and choose Rejoin call. Words spoken while disconnected will not be recovered.
• Calling not ready: use Check setup on the desktop Calls page, then contact us if the service remains unavailable.

Unopened invitations expire after 24 hours. The selected room timer starts when the room is first started, even if the other person arrives later. The service is currently configured for a maximum of 60 minutes per call; a closed or expired room needs a new invitation.

WhoSaid does not record call audio or save a call transcript in this version. The browser displays recent phrases; closing or reloading clears that page’s text. Rejoining resumes new captions.

Calls cost $0.03 USD per connected minute, counted once for the conversation while both people are present and prorated by the second. Call identifiers, language choices, connection times and duration are kept for support and billing. Provider data policies also apply.

The stated billing schedule is a separate charge on the 15th for the previous calendar month’s calls once usage billing is active. Usage is being recorded; automatic usage invoicing is not active yet. This is separate from the $1/hour rate for uploaded-file transcription.

1. Open Settings → WhoSaid updates and check Installed version.
2. Select Check for updates. If a release is available, choose Download update.
3. After verification, save your edits and finish active work. Choose Install update…, confirm, and follow the installer.
4. Choose Open WhoSaid on the installer’s final screen.

WhoSaid checks when opened and daily while running. You choose when to download and install. Updates preserve saved documents and your account connection. Normally, install over the existing app; uninstalling first is unnecessary.

Choose Settings → Manage account & billing, or open your account here. Use Payment & cancellation to update payment details or cancel.

The subscription is $4.99 USD per month and renews until cancelled. Cancel before the next renewal to avoid the next subscription charge. Access continues through the period already paid. Keep your saved files and backups.

Local CPU/GPU transcription is included. Optional translated calls cost $0.03 per connected minute; uploaded-file transcription costs $1 per recording hour. These usage services are separate from the subscription. See pricing and billing terms.

Local processing keeps the recording on your computer, but setup, account checks and updates need internet. Previously verified access can allow a limited offline period; indefinite offline access is not promised. Before working away from internet, finish model setup and select Check subscription while connected.

Online transcription, translated calls and signature delivery require internet. If account access fails, confirm the Windows date and time, sign in with the subscribed account and check again.

Show activity log at the bottom of the app opens progress and error details. Settings → Run Environment Check checks local tools. Appearance → Light / Dark changes the display and is remembered on this computer.

First-use downloads and CPU processing can take time. Open Show activity log to check progress and any reported error. Keep the computer awake and connected during setup or an online job. A short sample is a useful check before processing a long recording.

A two-hour recording may take two hours or longer on CPU. Medium needs more resources than smaller word-recognition models. Read the processing comparison before choosing a different option.

Speaker identification estimates voice similarity. Poor audio, brief replies, overlap and similar voices can confuse it. Longer recordings may provide more evidence, but do not guarantee accurate labels.

For local runs, use a realistic speaker range or the exact count if known. In the reader, give duplicate IDs the same actual name, and edit incorrectly assigned passages. See model settings and speaker naming.

• Recordings: Record and Transcribe → Open recordings folder.
• Transcripts: Transcribe → Open transcript folder, or Open Transcript Folder in the reader.
• Signed PDFs: Sign & Send → Open signed PDFs folder.
• Edited PDFs and videos: the folder you chose when saving or exporting; use Open folder after saving.

Back up important files using your normal firm process. Keep edited transcript TXT files beside their .txt.whosaid.json companions. WhoSaid’s temporary online delivery is not a permanent document backup.

Recording, CPU/GPU transcription, PDF editing and video/GIF exports run locally. Optional Deepgram file transcription uploads audio for processing. Sign & Send uploads a temporary signing copy. Calls send live audio for captions and translation.

Your account and billing checks also use the internet. “Local transcription” describes where your recording is processed; it does not mean the entire app never connects online. Review the upload or participant notice before using an online feature.

Return to your account and confirm you are signed in with the address used for the subscription. Allow the payment confirmation to finish, then refresh the page. In the app, use Connect account and Check subscription.

If subscription controls stay disabled after completing the required profile and agreeing to the terms, contact us with your account email and the visible message. Do not send card details or passwords. An invited client does not need a paid account to sign or join your call.

For an installed copy, open Windows Settings → Apps → Installed apps, find WhoSaid, and choose Uninstall. Use the current installer for updates instead of keeping several installed versions.

An older portable copy may not appear in Installed apps. Close that copy and remove its old shortcut or portable app folder only after checking that it does not contain recordings or documents you want to keep. Do not delete your transcript, recording, signed-PDF or speech-tools folders merely to remove a duplicate shortcut. Contact us if you are unsure which copy is running.