Still Guessing Who Said What? Split Speakers First

You have a multi-speaker recording. Two people talking. Maybe three. You need to know who said what.

Guessing from memory, or relabeling a transcript by hand, is the usual path. An audio diarization tool splits the file by speaker first so you stop guessing who said what. You still rename Speaker 1 and review the turns. Speaker diarization used to be a developer problem: call an API, write code, handle edge cases. For a non-technical business user, it was basically inaccessible.

Upload the file, split it by speaker, rename the labels, and review a timestamped transcript. You can also download isolated stems. No code required.

What an audio diarization tool returns

Speaker diarization is the process of segmenting an audio file by speaker identity. An audio diarization tool does that split and returns a labeled transcript. Isolated stems are a separate artifact: one audio file per voice, so you can hear each person on their own.

For the glossary and how the labeling step works, see the audio diarization tool guide. This page is the upload walkthrough.

The labeled transcript is typically something like:

[00:01:23] Speaker 1: We need to finalize the contract by Friday.
[00:01:28] Speaker 2: I'll have legal review it by Thursday.
[00:01:35] Speaker 1: Perfect. Let's confirm on the call next week.

People use that labeled first pass for:

• Meeting recordings that still say Speaker 1 / Speaker 2
• Interview recordings where you cannot tell who answered
• Calls you need to relabel before you share notes
• Podcasts with more than one voice, when you want labels and stems

This is a working labeled transcript. It is not a certified legal or medical record.

The Old Way (Painful)

Before no-code tools existed, you had two options:

1. Manual labeling — listen to the audio and type who said what as you go. Relabeling a wall of Speaker 1 / Speaker 2 by hand often takes longer than listening to the recording again.

2. Use an API — services like AssemblyAI, Deepgram, and AWS Transcribe all support diarization, but they require you to write code to call the API and process the output.

Neither is practical for most people who just need names on the turns.

The New Way: No-Code Speaker Separation

Speaker Separation lets you:

1. Upload an audio file
2. Optionally set the expected number of speakers (or leave it on auto)
3. Review a labeled, timestamped transcript and rename the speakers
4. Download isolated stems, or export txt / srt

That's it. No API keys. No code. No waiting for a developer.

Guest runs are free (2/day). Signed-in orgs pay 2 credits per minute, rounded up. You still rename Speaker 1 and review overlapping turns.

How It Works Under the Hood

Modern speaker separation uses neural diarization models that analyze voice characteristics — pitch, cadence, frequency patterns — to distinguish between speakers even when they're not explicitly identified by name. The audio diarization tool guide covers voice activity detection and speaker embeddings in more detail.

The split is harder when:

• Audio quality is phone-quality or a far-field room mic
• Speakers overlap — people talking over each other
• Many voices sound alike, or there are more than a few speakers

You still review those turns. Isolated stems help you hear one voice at a time. They do not certify the text.

Step-by-Step: Split Audio by Speaker

Step 1: Get Your Audio File

Supported formats are MP3, WAV, M4A, OGG, and WEBM. If your recording is a Zoom or Teams video, export the audio first, then upload that file.

Step 2: Upload to the Speaker Separation Tool

Go to AgentLeverage's speaker separation tool and upload your file. If you know how many speakers are in the recording, you can set that count — it can slightly improve the split.

Step 3: Review the Output

The tool returns a transcript segmented by speaker. Each segment is labeled Speaker 1, Speaker 2, and so on, with a timestamp. Rename the speakers after you review the split so the labels match the people on the call. Check the turns where people talked over each other.

Step 4: Export or Download Stems

Copy the labeled transcript, export it as text or SRT, or download isolated stems (one file per voice). That is a working labeled transcript, not a certified record.

Where a labeled split helps

A recording that still says Speaker 1 Zoom, Meet, and Teams exports often leave you guessing who said what. Label the voices, then rename Speaker 1.

Client or research interviews You get turns tagged by speaker so you can quote the right person. You still check the labels.

Calls you need to share A labeled transcript is easier to hand to someone who was not on the call. Isolated stems help when two voices overlap.

Podcasts and long recordings Labels and stems are a first pass for notes or an editor. We do not measure hours saved against a manual transcript.

Do not treat this as a deposition, court record, or compliance exhibit. If you need a certified copy, that still belongs to a court reporter or the process your matter already uses.

Alternatives to Consider

If you need speaker diarization at scale (hundreds of recordings per day), a direct API integration via AssemblyAI or Deepgram might fit a developer workflow better. For a file you already have, upload and review is the path that does not require writing code.

Other no-code options:

• Descript — editor plus transcription; check current pricing
• tl;dv — live meeting bot only, no file upload
• Otter.ai — transcription with some speaker ID; check the current free tier

Third-party prices change. We do not invent AgentLeverage ROI against those subscriptions.

For file-upload speaker separation, Speaker Separation is the live tool we ship.

If the recording already lands in n8n: run Speaker Separation, then Minutes from that job

If you already have Claude or ChatGPT open: the same sequence from a prompt

Frequently asked questions

What is an audio diarization tool?

Software that splits a multi-speaker recording so you can tell who said what. It labels Speaker 1 and Speaker 2, with timestamps. You rename those labels and review the transcript. It does not invent names from the audio. Isolated stems are optional extra files, one voice each. For the fuller glossary, see the audio diarization tool guide.

How is speaker diarization different from a plain transcript?

A plain transcript is a wall of text. Speaker diarization answers who spoke when. An audio diarization tool returns the labeled version so you are not reconstructing the conversation from memory.

Do I need to write code to split audio by speaker?

No. Upload the file to Speaker Separation, optionally set the speaker count, then review and rename. APIs from AssemblyAI, Deepgram, or AWS Transcribe still work if you want to build your own pipeline.

Can I try speaker separation without an account?

Yes. Guest runs are free and rate-limited (2/day). No credit card for a trial upload. Sign in if you want the job kept in a workspace.

What files can I upload?

MP3, WAV, M4A, OGG, and WEBM. If the source is a Zoom or Teams video, export the audio first, then upload that file.

When should I still label speakers by hand?

When people talk over each other a lot, the recording is phone-quality, or several voices sound alike. The tool still helps, but you should check those turns before you share the notes.

Is this a certified transcript?

No. It is a working labeled transcript. You still rename Speaker 1 and review overlaps. It is not a court record, not a medical record, and not something you file as the official copy.

Try Speaker Separation

Upload a recording. Review the labeled speaker output. Rename Speaker 1.

Try Speaker Separation →

Guest runs are free (2/day). No credit card. No code. You still rename and review.