End the Zoom meeting and let the recording process

Click End Meeting for All and wait until Zoom finishes processing the recording before renaming, moving, or uploading any files.

Local recordings are converted on the host computer, while cloud recordings are processed on Zoom's servers. Closing the application or shutting down the host computer too early can interrupt local conversion.

For a local recording, Zoom normally opens the completed recording folder after conversion. Processing can take several minutes for a long meeting, and high-resolution video or multiple recording layouts can increase the processing time.

Locate a local Zoom recording

Find a local recording on the computer used by the host or designated recorder, beginning with Zoom's configured recording directory and the standard Documents folder.

Each meeting usually has its own folder named with the date, time, and meeting topic.

Default Zoom recording locations

Zoom users can change this location. To find the configured folder:

1. Open the Zoom Workplace desktop application.
2. Click the profile image or settings icon.
3. Open Settings.
4. Select Recording.
5. Find the local recording location.
6. Click Open to display the folder.

Zoom changes application labels between releases, but the recording path remains available in the desktop application's recording settings.

Which local file should you use?

Start with audio_only.m4a. It contains the meeting audio without video data, so it transfers faster and consumes less storage. Upload zoom_0.mp4 when the M4A file is missing or visual review will help identify speakers.

Use separate participant audio files whenever Zoom created them. Isolated speaker tracks reduce overlap and provide cleaner material for speaker labeling.

Download a Zoom cloud recording

Open the recording in the Zoom web portal and download its audio-only M4A or MP4 file to an access-controlled project folder.

Cloud recordings are available to the recording owner and authorized administrators. If the owner disabled downloads, request direct access rather than recording browser playback through another application.

1. Sign in to the Zoom web portal.
2. Open Recordings & Transcripts.
3. Select Cloud Recordings.
4. Search by meeting topic, meeting ID, host, or date.
5. Open the correct meeting.
6. Review the available recording files.
7. Download the audio-only M4A file or the MP4 video.
8. Save the file to an access-controlled project folder.

A cloud recording can contain active-speaker, gallery-view, and shared-screen layouts. These versions normally contain the same conversation, so downloading every layout wastes storage. Choose audio-only unless the review process requires visual context.

Cloud processing can take longer for lengthy meetings. If the status remains Processing, wait for Zoom's completion email and refresh the portal. If a recording has disappeared, check the account's retention policy and trash area before assuming it was deleted permanently.

Extract audio from the Zoom video when required

Manual extraction is normally unnecessary because SpeechText.AI accepts common audio and video formats; use FFmpeg only when file size, codec compatibility, or an internal media policy requires separate audio.

Copy the original audio without re-encoding

Install FFmpeg, open a terminal in the recording folder, and run:

ffmpeg -i "zoom_0.mp4" -map 0:a:0 -c:a copy "meeting-audio.m4a"

This command:

• Reads zoom_0.mp4
• Selects the first audio stream
• Copies the existing audio without recompression
• Saves the stream in an M4A container

Copying the stream preserves the source quality and runs faster than encoding a new file.

Create a standard WAV file

If the transcription system or internal process requires WAV, use:

ffmpeg -i "zoom_0.mp4" -map 0:a:0 -ac 1 -ar 16000 -c:a pcm_s16le "meeting-audio.wav"

This creates a mono, 16 kHz, 16-bit PCM file. The format works well for speech recognition, but it is much larger than M4A. Do not convert to WAV simply because the format appears more professional: conversion cannot restore detail the original Zoom recording never captured.

ffprobe -v error \
-select_streams a:0 \
-show_entries stream=codec_name,sample_rate,channels,duration \
-of default=noprint_wrappers=1 \
"audio_only.m4a"

Upload the recording to SpeechText.AI

Upload the original M4A or MP4, choose the correct language and domain model, add relevant corporate terminology, and enable speaker recognition before starting transcription.

This workflow gives teams direct control over jargon, speaker labels, and output quality before the transcript enters company systems.

1. Sign in to SpeechText.AI and create a transcription task.
2. Upload audio_only.m4a, meeting-audio.m4a, or the original Zoom MP4.
3. Select the meeting's primary language and dialect.
4. Choose the domain model that matches the subject matter.
5. Add a custom vocabulary containing internal terminology.
6. Activate speaker diarization for mixed meeting audio.
7. Keep separate channels or participant tracks separate where available.
8. Start transcription.
9. Open the result in the transcript editor.

SpeechText.AI is designed for post-meeting workflows in which teams repeatedly discuss specialized products, clients, systems, and internal processes. Domain-specific models and custom vocabulary help with terms that general captions can miss. Multi-channel processing is also useful for interviews, support calls, panel discussions, and conference sessions.

Build a useful custom vocabulary

Do not upload an entire corporate dictionary to every job. Add terms that are likely to appear in the meeting and terms that recognition systems commonly confuse.

SpeechText.AI
QBR
NetSuite
Ceryx
AcuForm
CX-410
NPS
SOC 2
Aleksandra Nowak
Project Northstar

Include official spelling, capitalization, spaces, and punctuation. For acronyms, note how employees pronounce them. "NPS" spoken as three letters differs acoustically from an acronym pronounced as a single word.

Create reusable vocabulary sets for departments such as engineering, finance, legal, healthcare, customer support, and product management. Update the relevant set after transcript review: every corrected proper noun is a candidate for the next meeting's vocabulary list.

Review speakers, timestamps, and terminology

Review the transcript against the recording, prioritizing participant names, numbers, decisions, action items, repeated terminology, overlapping speech, and source timestamps.

1. Confirm the meeting identity. Check the title, date, participants, and recording duration.
2. Rename speakers. Replace labels such as "Speaker 1" with confirmed names or roles.
3. Check custom terms. Search for product names, client names, acronyms, and technical phrases.
4. Verify numbers. Review prices, dates, percentages, ticket numbers, and version identifiers.
5. Inspect overlapping speech. Listen to sections where two people interrupted each other.
6. Correct punctuation. Split long speech blocks into readable sentences and paragraphs.
7. Preserve useful timestamps. Keep timestamps beside decisions, commitments, and disputed statements.
8. Export the approved version. Choose a text, document, or caption format that fits the next task.

Keep the original recording and approved transcript linked through the same meeting ID or file-naming convention.

Turn the transcript into an actionable meeting record

Keep the full transcript as the source record, then place a concise meeting brief above it that identifies decisions, action owners, deadlines, open questions, risks, and supporting timestamps.

Use this structure:

# Meeting title

Date:
Host:
Participants:
Recording ID:

## Decisions
- Decision:
- Approved by:
- Timestamp:

## Action items
| Action | Owner | Deadline | Transcript timestamp |
|---|---|---|---|

## Open questions
- Question:
- Assigned to:
- Review date:

## Risks and blockers
- Risk:
- Impact:
- Owner:

## Full transcript
[Speaker-labeled transcript with timestamps]

Timestamps make the document auditable. If someone disputes an assignment or deadline, the reader can return to the precise point in the recording instead of searching through an hour of video.

Store the action summary and transcript together. A summary without its source loses context, while a transcript without extracted actions is more likely to be ignored.