[Integrations](https://spiral.ai/integrations) › [Hermes](https://spiral.ai/integrations/hermes) › [SAM Audio](https://spiral.ai/integrations/hermes/sam-audio)

# How to Use SAM Audio in Hermes

Use SAM Audio to extract speech, instruments, or sound effects from a mixed recording by describing the sound. It returns the requested sound and a separate track containing everything else.

|  |  |
| --- | --- |
| Model | SAM Audio |
| Available through | fal |
| Category | Audio |
| Price | $0.065 per 30 seconds of audio, at least 30 seconds. Each extra reranking candidate: $0.0325 per 30 seconds. |
| Returns | audio |
| Runs | In the background |
| Setup | About 2 minutes, no API key |

## Prompts to try with SAM Audio in Hermes

- Every weekday, use SAM Audio to extract 'speech' from newly submitted interview recordings. Keep both returned tracks with each recording for editorial review.
- Each evening, use SAM Audio on new podcast recordings to extract 'dog barking' with automatic sound timing enabled. Collect the remaining-audio tracks for review before publication.
- Every Monday, use SAM Audio to extract 'piano playing' from newly submitted rehearsal recordings. Leave automatic sound timing off and collect the piano tracks for the music coach.
- Each morning, use SAM Audio to extract 'car honking' from new location recordings. Enable automatic detection of when the sound occurs and retain both returned tracks.
- After each weekly recording session, use SAM Audio to extract 'singing voice' from the submitted song recordings. Generate and rank three candidates per recording and collect the selected results for listening.

## How to set up SAM Audio in Hermes

Pick the way that suits you. Each takes about two minutes.

### Use the Spiral skill

1. **Send Hermes this message**: Your agent reads Spiral's skill and sets itself up.

```
set up https://spiral.ai/SKILL.md
```

2. **Sign in or paste a key**: Approve Spiral in the browser, or give the agent a key from the console.

### Use Spiral's MCP server

1. **Add the server and sign in**: Hermes opens your browser: sign in to Spiral and click Authorize. No API key. When Hermes asks Enable all tools? \[Y/n/select], answer Y: n cancels without saving the server, and select hides the tools you leave out.

```bash
hermes mcp add spiral \
  --url https://api.spiral.ai/mcp \
  --auth oauth \
  --connect-timeout 300
```

2. **Check it**: In a chat that was already open, run `/reload-mcp`.

```bash
hermes mcp list
hermes mcp test spiral
```

### Use the Spiral CLI

1. **Install and sign in**: Installs the Spiral CLI, signs you in and adds its skill to your coding agents.

```bash
npx -y spiral-cli@latest init
```

### Call the Spiral API

1. **Create an API key**: In the console, under API keys.

2. **Run SAM Audio**: One request; the result comes back by polling or a webhook.

```bash
curl https://api.spiral.ai/v1/runs \
  -H "Authorization: Bearer $SPIRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
    "provider": "fal",
    "model": "fal-ai/sam-audio/separate",
    "input": {
      "audio_url": "https://v3b.fal.media/files/b/0a88511f/tUUCI9eDmeC2RqJEOXrZk_assets_avatar_multi_sing_woman.WAV",
      "prompt": "piano playing"
    }
  }'
```

Then paste one of the prompts above: Hermes finds SAM Audio on Spiral, shows the price and runs it.

Setup guide in the docs: https://spiral.ai/docs/quickstarts/hermes

## SAM Audio tools and prices

- [SAM Audio · Text-Prompted Sound Separation](https://spiral.ai/tools/fal/fal-ai/sam-audio/separate) (fal): $0.065 per 30 seconds of audio, at least 30 seconds. Each extra reranking candidate: $0.0325 per 30 seconds.

## Settings you can ask for

| Setting | Allowed values | Default | Description |
| --- | --- | --- | --- |
| predict_spans |  | false | Automatically predict temporal spans where the target sound occurs. |
| reranking_candidates | 1 to 7 | 1 | Number of candidates to generate and rank. Higher improves quality but increases latency and cost. |
| acceleration | fast, balanced, quality | balanced | The acceleration level to use. |
| max_chunk_duration | 10 to 60 | 60 | Maximum audio duration (in seconds) to process in a single pass. Longer audio will be chunked with overlap and blended. |
| chunk_overlap | 0 to 30 | 5 | Overlap duration (in seconds) between chunks for crossfade blending. |
| output_format | wav, mp3 | wav | Output audio format. |

## What can you do with SAM Audio in Hermes?

Prepare described sound separations from recurring recordings before an editor reviews them. A scheduled agent can apply consistent sound descriptions and collect extracted and remaining-audio tracks for each recording.

### Prepare interview dialogue

Extract speech from a noisy interview recording for a dialogue edit. Compare the extracted speech with the original before replacing the source track.

### Remove barking from a podcast

Target the barking and use the remaining-audio track as a podcast editing candidate. Check the moments where barking overlaps speech.

### Reduce traffic in outdoor recordings

Separate traffic noise from recorded outdoor audio, then review the remaining track for use beneath the scene.

### Study a piano part

Extract piano from a mixed music recording to hear its phrasing more clearly. Treat the result as a listening aid and review passages with overlapping instruments.

### Prepare a vocal editing reference

Separate singing from accompaniment for an editing reference. Avoid relying on it to identify one singer within a chorus.

### Collect recorded sound effects

Extract a specific event, such as an engine revving or a horn honking, from a location recording for a sound editing session.

### Separate music from spoken content

Target background music in a mixed recording and review the remaining track for a speech-focused edit.

## Overview

- [fal](https://spiral.ai/providers/fal)
- [All Hermes guides](https://spiral.ai/integrations/hermes)

## SAM Audio in other agents

- [Amp](https://spiral.ai/integrations/amp/sam-audio)
- [ChatGPT](https://spiral.ai/integrations/chatgpt/sam-audio)
- [Claude](https://spiral.ai/integrations/claude-desktop/sam-audio)
- [Claude Code](https://spiral.ai/integrations/claude-code/sam-audio)
- [Cline](https://spiral.ai/integrations/cline/sam-audio)
- [Codex](https://spiral.ai/integrations/codex/sam-audio)
- [Cursor](https://spiral.ai/integrations/cursor/sam-audio)
- [Devin](https://spiral.ai/integrations/devin/sam-audio)
- [dots](https://spiral.ai/integrations/dots/sam-audio)
- [Factory Droid](https://spiral.ai/integrations/factory-droid/sam-audio)
- [Gemini](https://spiral.ai/integrations/gemini/sam-audio)
- [Gemini CLI](https://spiral.ai/integrations/gemini-cli/sam-audio)
- [Google Antigravity](https://spiral.ai/integrations/antigravity/sam-audio)
- [goose](https://spiral.ai/integrations/goose/sam-audio)
- [Grok](https://spiral.ai/integrations/grok/sam-audio)
- [Grok Build](https://spiral.ai/integrations/grok-build/sam-audio)
- [Instinct](https://spiral.ai/integrations/instinct/sam-audio)
- [Manus](https://spiral.ai/integrations/manus/sam-audio)
- [Muse](https://spiral.ai/integrations/muse/sam-audio)
- [Muse Code](https://spiral.ai/integrations/muse-code/sam-audio)
- [n8n](https://spiral.ai/integrations/n8n/sam-audio)
- [Notion](https://spiral.ai/integrations/notion/sam-audio)
- [OpenClaw](https://spiral.ai/integrations/openclaw/sam-audio)
- [opencode](https://spiral.ai/integrations/opencode/sam-audio)
- [Perplexity](https://spiral.ai/integrations/perplexity/sam-audio)
- [Raycast](https://spiral.ai/integrations/raycast/sam-audio)
- [VS Code](https://spiral.ai/integrations/vscode/sam-audio)
- [Windsurf](https://spiral.ai/integrations/windsurf/sam-audio)
- [Zapier](https://spiral.ai/integrations/zapier/sam-audio)

## More Hermes guides

- [Qwen3-TTS 1.7B Voice Design](https://spiral.ai/integrations/hermes/qwen3-tts-1-7b-voice-design)
- [SAM Audio Visual Separation](https://spiral.ai/integrations/hermes/sam-audio-visual-separation)
- [Qwen3-TTS 1.7B](https://spiral.ai/integrations/hermes/qwen3-tts-1-7b)
- [Scribe v2](https://spiral.ai/integrations/hermes/scribe-v2)
- [Qwen3-TTS 0.6B Clone Voice](https://spiral.ai/integrations/hermes/qwen3-tts-0-6b-clone-voice)
- [Scribe v3](https://spiral.ai/integrations/hermes/scribe-v3)

## Questions about SAM Audio in Hermes

**Can Hermes use SAM Audio?**

Yes. Add Spiral to Hermes as an MCP server, and Hermes can find and run SAM Audio like any built-in tool.

**How much does SAM Audio cost in Hermes?**

$0.065 per 30 seconds of audio, at least 30 seconds. Each extra reranking candidate: $0.0325 per 30 seconds. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Hermes plan is separate.

**Do I need a fal account to use SAM Audio in Hermes?**

No. Spiral runs SAM Audio on its own fal account. You sign in to Spiral once and never paste a key.

**How long does it take to set up SAM Audio in Hermes?**

About two minutes: add the server and sign in, then check it.

**Does Hermes wait while SAM Audio runs?**

SAM Audio runs in the background. Hermes starts the run, checks on it and gives you the audio when it's ready.

**What is SAM Audio useful for?**

SAM Audio extracts a described sound from a mixed recording. Use it to isolate speech, music, or everyday sound effects for editing and review.

**How should I describe the sound I want?**

Use a short, lowercase phrase naming the source and its action, such as “piano playing” or “dog barking.” Meta recommends noun or verb phrases that match how the model was trained.

**Can SAM Audio remove a sound instead of keeping it?**

Yes. Describe the unwanted sound, then use the remaining-audio track, called the residual, for the edit.

**What sounds are hardest to separate?**

Highly similar sources remain difficult, such as one singer within a chorus or one instrument within an orchestra. Review those results before using them in a finished edit.

**When should I try automatic sound timing?**

Try it for distinct events, such as barking or honking, so the model predicts when the target occurs. Compare results with it disabled for continuous instrument parts: Meta reported a small decline in professional instrument separation.

