{"slug":"merlino-voice","title":"Merlino Voice — Fish Audio Cloned Voice for Outbound TTS","tags":["merlino-voice","fish-audio","tts","voice-clone","outbound-voice","signalwire"],"agent_summary":"Merlino Voice is Mike's custom-cloned voice on Fish Audio used for branded outbound TTS — agent introductions, voicemail drops, IVR greetings, ringless drops. Distinct from generic Fish Audio TTS because the voice model is identity-locked. Voice ID, default parameters (speed 1.0, energy 0.7), audio formats (mp3 16-bit, opus, wav), and SignalWire integration via SWML play verb.","trigger_phrases":["Merlino voice","Mike Merlino cloned voice","branded TTS voice","agency voice clone","personal voice clone fish audio","Merlino TTS for calls"],"runnable":true,"markdown":"\n# Merlino Voice\n\nMerlino Voice is Mike Merlino's custom-cloned voice on Fish Audio. Distinct from generic Fish Audio TTS because the voice model is identity-locked — it sounds like Mike. Used for branded agency communications where personal touch matters: outbound voicemails, video intros, podcast intros, branded IVR greetings, sales agent self-introductions.\n\nFor platform-level Fish Audio reference (auth, API surface, model selection), see [Fish Audio TTS](/topic/fish-audio-tts).\n\n## When to use Merlino Voice vs generic TTS\n\n| Use case | Voice |\n|---|---|\n| Generic IVR (\"press 1 for sales\") | Generic ElevenLabs/Deepgram |\n| Mike's personal cold-call voicemail drop | Merlino Voice |\n| Agency podcast intro | Merlino Voice |\n| Client-specific AI receptionist | Custom voice per client |\n| Mike speaking at an event (recorded TTS) | Merlino Voice |\n| Mass automated alerts | Generic TTS (cheaper, more neutral) |\n\nCloned voice carries a personal-brand signature. Use sparingly — overuse dilutes the \"this is from Mike personally\" effect.\n\n## Configuration\n\n| Parameter | Value |\n|---|---|\n| Voice ID | (Fish Audio model ID — kept in `D:/Ecosystem/secrets/MASTER_API_KEYS.env`) |\n| Default speed | 1.0 |\n| Default energy | 0.7 |\n| Default chunk length | 200 |\n| Format | mp3 (16-bit) for general use, wav for SignalWire `play`, opus for streaming |\n\nAPI key lives at `FISH_AUDIO_API_KEY` in the secrets file.\n\n## Generation example\n\n```python\nimport os\nimport requests\n\nFISH_API = \"https://api.fish.audio/v1/tts\"\nFISH_KEY = os.environ[\"FISH_AUDIO_API_KEY\"]\nMERLINO_VOICE_ID = os.environ[\"MERLINO_VOICE_ID\"]\n\ndef generate_merlino_audio(text, output_path, format=\"mp3\", speed=1.0, energy=0.7):\n response = requests.post(\n FISH_API,\n headers={\n \"Authorization\": f\"Bearer {FISH_KEY}\",\n \"Content-Type\": \"application/json\",\n },\n json={\n \"text\": text,\n \"reference_id\": MERLINO_VOICE_ID,\n \"format\": format,\n \"mp3_bitrate\": 128,\n \"chunk_length\": 200,\n \"normalize\": True,\n \"latency\": \"balanced\", # or \"normal\" for higher quality, slower\n }\n )\n response.raise_for_status()\n with open(output_path, \"wb\") as f:\n f.write(response.content)\n return output_path\n```\n\nTypical generation time: 800ms - 2s for short utterances under 200 characters, 3-8s for paragraph-length text.\n\n## SignalWire integration via `play`\n\nFor SWML scripts that need a Merlino-voiced introduction or message, pre-generate the audio and host it, then reference via the SWML `play` verb:\n\n```yaml\nversion: 1.0.0\nsections:\n main:\n - answer: {}\n - play:\n url: https://cdn.merlinoai.com/voice/intro-mike-v3.mp3\n - connect:\n to: sip:agent@pbx.merlinoai.com\n```\n\nPre-generating is better than on-the-fly generation inside the call because:\n\n- Eliminates Fish Audio API latency from the call path\n- Allows audio quality tuning (multiple takes, normalization)\n- CDN delivery is faster than re-rendering\n\nFor dynamic per-call content (using caller's name, etc.), use Fish Audio's WebSocket streaming endpoint and tap output directly into SignalWire — but this adds 1-2 seconds of perceived latency.\n\n## Voicemail drop with Merlino Voice\n\nA signature use case. Mike's personal-sounding voicemail beats generic dialer voicemails.\n\n```yaml\n# voicemail-drop-merlino.yaml\nversion: 1.0.0\nsections:\n main:\n - answer:\n answer_on: machine_end_beep\n - play:\n url: https://cdn.merlinoai.com/voice/cold-drop-v7.mp3\n - hangup: {}\n```\n\nOutbound call with AMD targeting voicemail:\n\n```python\nclient.calls.create(\n to=lead.phone,\n from_=OUTBOUND_DID,\n url=\"https://your.api/voicemail-drop-merlino.xml\",\n machine_detection=\"DetectMessageEnd\",\n)\n```\n\nSee [voicemail drop](/topic/voicemail-drop) for the full pattern. Cloned-voice voicemail drops require even tighter TCPA discipline because they're personal-sounding — the recipient assumes a real person and reacts more strongly when they discover it's pre-recorded.\n\n## Generation patterns\n\n### Pattern 1: Pre-render variations to CDN\n\nCommon patterns rendered ahead of time and uploaded to CDN:\n\n```bash\ntexts=(\n \"intro-mike-v1:Hey, this is Mike Merlino. Just calling about your business.\"\n \"intro-mike-v2:Hi, it's Mike at Merlino AI. Quick question for you.\"\n \"vmdrop-mike-v1:Hey, Mike Merlino here. Missed you — give me a call back when you have a sec.\"\n)\n\nfor entry in \"${texts[@]}\"; do\n name=\"${entry%%:*}\"\n text=\"${entry#*:}\"\n python generate.py --text \"$text\" --out \"/tmp/$name.mp3\"\n aws s3 cp \"/tmp/$name.mp3\" \"s3://cdn-merlinoai/voice/$name.mp3\" --acl public-read\ndone\n```\n\n### Pattern 2: Personalized per-recipient render\n\nWhen the message needs to address the recipient by name:\n\n```python\ndef generate_personalized(recipient_first_name, output_path):\n text = f\"Hey {recipient_first_name}, it's Mike. Got a sec to talk?\"\n return generate_merlino_audio(text, output_path)\n```\n\nCache renders by text-content hash to avoid re-generating identical messages.\n\n### Pattern 3: Live during-call generation (advanced)\n\nFor dynamic content during an active call, use the AI verb with Fish Audio configured as the TTS provider:\n\n```yaml\n- ai:\n prompt:\n text: \"You are Mike Merlino, agency owner. Be friendly and direct.\"\n languages:\n - name: English (Merlino)\n code: en\n voice: fishaudio.${MERLINO_VOICE_ID}\n engine: fishaudio\n SWAIG:\n functions: []\n```\n\nSignalWire's AI verb supports Fish Audio as a TTS engine. Latency is higher than ElevenLabs/Deepgram. Best when the personal voice signature is more important than perfect responsiveness.\n\n## Quality tuning\n\n| Issue | Cause | Fix |\n|---|---|---|\n| Robotic cadence | `chunk_length` too short | Increase to 200-300 |\n| Word emphasis wrong | Missing punctuation hints | Add commas, em-dashes for natural pauses |\n| Volume too low for phone playback | TTS output is line-level, phone expects -6 to -3 dB | Normalize with `ffmpeg -af \"loudnorm=I=-16:LRA=11:TP=-1.5\"` |\n| Sibilance harsh | Source recording had it | De-ess with `ffmpeg -af \"highshelf=f=6000:g=-3\"` |\n| Mismatched pitch across renders | Energy/speed varied between sessions | Lock parameters in a single config |\n\n## Audio post-processing for telephony\n\nPhone networks downsample to 8 kHz. To make Merlino Voice sound consistent over phone:\n\n```bash\n# Render at 24 kHz\nfish_audio_render --voice $MERLINO_ID --text \"...\" --format wav -o raw.wav\n\n# Process for telephony\nffmpeg -i raw.wav \\\n -af \"loudnorm=I=-16:LRA=11:TP=-1.5,highshelf=f=6000:g=-3\" \\\n -ar 8000 -ac 1 -acodec pcm_mulaw \\\n telephony.wav\n```\n\nResult is 8 kHz mono mulaw, the exact format G.711 carriers use. Plays without resampling latency.\n\n## Cost\n\nFish Audio TTS pricing (as of 2025-11):\n\n| Item | Cost |\n|---|---|\n| Per character | ~$0.000015 |\n| Per second of audio | ~$0.0008 |\n| Voice clone training (one-time) | $50-200 depending on tier |\n| Custom voice retention | Included in plan |\n\nA 30-second cold-voicemail message: ~$0.025 per render. Cache aggressively.\n\n## Voice integrity guardrails\n\nThe Merlino Voice clone is a brand asset. Misuse damages the brand. Operational rules:\n\n1. **Never use for impersonation** — voice clones generating content \"as Mike\" without his approval is a hard no.\n2. **Audit log every render** — `D:/Ecosystem/logs/merlino-voice-renders.log` records every API call with text content and use case.\n3. **Approved sequences only** — production sequences using Merlino Voice are version-controlled in `D:/Ecosystem/voice-scripts/`.\n4. **Time-bounded access** — staff API key access to the Merlino voice ID has a clear approval chain.\n\n## Common pitfalls\n\n- **Generating on-the-fly during calls** — adds 2-5 second latency. Pre-render to CDN.\n- **No normalization on phone audio** — Merlino Voice generated at -23 LUFS sounds quiet on phone. Always loudnorm to -16 LUFS.\n- **Calling Fish Audio from inside SWML directly** — works but not always reliable. Pre-render or use the AI verb with Fish Audio engine.\n- **Misusing for marketing voicemail without consent** — see [voicemail drop](/topic/voicemail-drop). TCPA is strict; cloned voice raises the stakes.\n- **Voice drift over Fish Audio version updates** — Fish Audio occasionally updates models. Periodically re-render reference samples and compare to ensure consistency.\n\n## Related patterns\n\n- [Fish Audio TTS](/topic/fish-audio-tts) — generic Fish Audio API reference\n- [Voicemail drop](/topic/voicemail-drop) — pre-recorded outbound voicemail\n- [SignalWire AI receptionist](/topic/signalwire-ai-receptionist) — AI voice agent with custom TTS\n\n## References\n\n- Fish Audio API documentation\n- SignalWire SWML `play` verb\n- SignalWire AI verb language/voice configuration\n","html":"

Merlino Voice

\n

Merlino Voice is Mike Merlino's custom-cloned voice on Fish Audio. Distinct from generic Fish Audio TTS because the voice model is identity-locked — it sounds like Mike. Used for branded agency communications where personal touch matters: outbound voicemails, video intros, podcast intros, branded IVR greetings, sales agent self-introductions.

\n

For platform-level Fish Audio reference (auth, API surface, model selection), see Fish Audio TTS.

\n

When to use Merlino Voice vs generic TTS

\n

| Use case | Voice |\n|---|---|\n| Generic IVR (\"press 1 for sales\") | Generic ElevenLabs/Deepgram |\n| Mike's personal cold-call voicemail drop | Merlino Voice |\n| Agency podcast intro | Merlino Voice |\n| Client-specific AI receptionist | Custom voice per client |\n| Mike speaking at an event (recorded TTS) | Merlino Voice |\n| Mass automated alerts | Generic TTS (cheaper, more neutral) |

\n

Cloned voice carries a personal-brand signature. Use sparingly — overuse dilutes the \"this is from Mike personally\" effect.

\n

Configuration

\n

| Parameter | Value |\n|---|---|\n| Voice ID | (Fish Audio model ID — kept in D:/Ecosystem/secrets/MASTER_API_KEYS.env) |\n| Default speed | 1.0 |\n| Default energy | 0.7 |\n| Default chunk length | 200 |\n| Format | mp3 (16-bit) for general use, wav for SignalWire play, opus for streaming |

\n

API key lives at FISH_AUDIO_API_KEY in the secrets file.

\n

Generation example

\n
import os\nimport requests\n\nFISH_API = \"https://api.fish.audio/v1/tts\"\nFISH_KEY = os.environ[\"FISH_AUDIO_API_KEY\"]\nMERLINO_VOICE_ID = os.environ[\"MERLINO_VOICE_ID\"]\n\ndef generate_merlino_audio(text, output_path, format=\"mp3\", speed=1.0, energy=0.7):\n    response = requests.post(\n        FISH_API,\n        headers={\n            \"Authorization\": f\"Bearer {FISH_KEY}\",\n            \"Content-Type\": \"application/json\",\n        },\n        json={\n            \"text\": text,\n            \"reference_id\": MERLINO_VOICE_ID,\n            \"format\": format,\n            \"mp3_bitrate\": 128,\n            \"chunk_length\": 200,\n            \"normalize\": True,\n            \"latency\": \"balanced\",  # or \"normal\" for higher quality, slower\n        }\n    )\n    response.raise_for_status()\n    with open(output_path, \"wb\") as f:\n        f.write(response.content)\n    return output_path\n
\n

Typical generation time: 800ms - 2s for short utterances under 200 characters, 3-8s for paragraph-length text.

\n

SignalWire integration via play

\n

For SWML scripts that need a Merlino-voiced introduction or message, pre-generate the audio and host it, then reference via the SWML play verb:

\n
version: 1.0.0\nsections:\n  main:\n    - answer: {}\n    - play:\n        url: https://cdn.merlinoai.com/voice/intro-mike-v3.mp3\n    - connect:\n        to: sip:agent@pbx.merlinoai.com\n
\n

Pre-generating is better than on-the-fly generation inside the call because:

\n\n

For dynamic per-call content (using caller's name, etc.), use Fish Audio's WebSocket streaming endpoint and tap output directly into SignalWire — but this adds 1-2 seconds of perceived latency.

\n

Voicemail drop with Merlino Voice

\n

A signature use case. Mike's personal-sounding voicemail beats generic dialer voicemails.

\n
# voicemail-drop-merlino.yaml\nversion: 1.0.0\nsections:\n  main:\n    - answer:\n        answer_on: machine_end_beep\n    - play:\n        url: https://cdn.merlinoai.com/voice/cold-drop-v7.mp3\n    - hangup: {}\n
\n

Outbound call with AMD targeting voicemail:

\n
client.calls.create(\n    to=lead.phone,\n    from_=OUTBOUND_DID,\n    url=\"https://your.api/voicemail-drop-merlino.xml\",\n    machine_detection=\"DetectMessageEnd\",\n)\n
\n

See voicemail drop for the full pattern. Cloned-voice voicemail drops require even tighter TCPA discipline because they're personal-sounding — the recipient assumes a real person and reacts more strongly when they discover it's pre-recorded.

\n

Generation patterns

\n

Pattern 1: Pre-render variations to CDN

\n

Common patterns rendered ahead of time and uploaded to CDN:

\n
texts=(\n    \"intro-mike-v1:Hey, this is Mike Merlino. Just calling about your business.\"\n    \"intro-mike-v2:Hi, it's Mike at Merlino AI. Quick question for you.\"\n    \"vmdrop-mike-v1:Hey, Mike Merlino here. Missed you — give me a call back when you have a sec.\"\n)\n\nfor entry in \"${texts[@]}\"; do\n    name=\"${entry%%:*}\"\n    text=\"${entry#*:}\"\n    python generate.py --text \"$text\" --out \"/tmp/$name.mp3\"\n    aws s3 cp \"/tmp/$name.mp3\" \"s3://cdn-merlinoai/voice/$name.mp3\" --acl public-read\ndone\n
\n

Pattern 2: Personalized per-recipient render

\n

When the message needs to address the recipient by name:

\n
def generate_personalized(recipient_first_name, output_path):\n    text = f\"Hey {recipient_first_name}, it's Mike. Got a sec to talk?\"\n    return generate_merlino_audio(text, output_path)\n
\n

Cache renders by text-content hash to avoid re-generating identical messages.

\n

Pattern 3: Live during-call generation (advanced)

\n

For dynamic content during an active call, use the AI verb with Fish Audio configured as the TTS provider:

\n
- ai:\n    prompt:\n      text: \"You are Mike Merlino, agency owner. Be friendly and direct.\"\n    languages:\n      - name: English (Merlino)\n        code: en\n        voice: fishaudio.${MERLINO_VOICE_ID}\n        engine: fishaudio\n    SWAIG:\n      functions: []\n
\n

SignalWire's AI verb supports Fish Audio as a TTS engine. Latency is higher than ElevenLabs/Deepgram. Best when the personal voice signature is more important than perfect responsiveness.

\n

Quality tuning

\n

| Issue | Cause | Fix |\n|---|---|---|\n| Robotic cadence | chunk_length too short | Increase to 200-300 |\n| Word emphasis wrong | Missing punctuation hints | Add commas, em-dashes for natural pauses |\n| Volume too low for phone playback | TTS output is line-level, phone expects -6 to -3 dB | Normalize with ffmpeg -af \"loudnorm=I=-16:LRA=11:TP=-1.5\" |\n| Sibilance harsh | Source recording had it | De-ess with ffmpeg -af \"highshelf=f=6000:g=-3\" |\n| Mismatched pitch across renders | Energy/speed varied between sessions | Lock parameters in a single config |

\n

Audio post-processing for telephony

\n

Phone networks downsample to 8 kHz. To make Merlino Voice sound consistent over phone:

\n
# Render at 24 kHz\nfish_audio_render --voice $MERLINO_ID --text \"...\" --format wav -o raw.wav\n\n# Process for telephony\nffmpeg -i raw.wav \\\n    -af \"loudnorm=I=-16:LRA=11:TP=-1.5,highshelf=f=6000:g=-3\" \\\n    -ar 8000 -ac 1 -acodec pcm_mulaw \\\n    telephony.wav\n
\n

Result is 8 kHz mono mulaw, the exact format G.711 carriers use. Plays without resampling latency.

\n

Cost

\n

Fish Audio TTS pricing (as of 2025-11):

\n

| Item | Cost |\n|---|---|\n| Per character | ~$0.000015 |\n| Per second of audio | ~$0.0008 |\n| Voice clone training (one-time) | $50-200 depending on tier |\n| Custom voice retention | Included in plan |

\n

A 30-second cold-voicemail message: ~$0.025 per render. Cache aggressively.

\n

Voice integrity guardrails

\n

The Merlino Voice clone is a brand asset. Misuse damages the brand. Operational rules:

\n
    \n
  1. Never use for impersonation — voice clones generating content \"as Mike\" without his approval is a hard no.
    2. \n
  2. Audit log every render — D:/Ecosystem/logs/merlino-voice-renders.log records every API call with text content and use case.
    3. \n
  3. Approved sequences only — production sequences using Merlino Voice are version-controlled in D:/Ecosystem/voice-scripts/.
    4. \n
  4. Time-bounded access — staff API key access to the Merlino voice ID has a clear approval chain.
    5. \n
\n

Common pitfalls

\n\n

Related patterns

\n\n

References

\n\n"}