Speak Session Enhancement

Scenario

Enhance Mavera Speak sessions by assigning custom ElevenLabs voices to different personas. Each persona gets a unique voice, creating an immersive multi-voice audio experience.

Flow: Mavera GET /personas → Match to ElevenLabs voices → POST /mave/chat (in-character) → ElevenLabs TTS per persona → Multi-voice audio

Code

Python

import os, requests, time
EL_KEY = os.environ["ELEVENLABS_API_KEY"]
EL_BASE = "https://api.elevenlabs.io/v1"
MV = os.environ["MAVERA_API_KEY"]
MV_BASE = "https://app.mavera.io/api/v1"
MV_H = {"Authorization": f"Bearer {MV}", "Content-Type": "application/json"}
os.makedirs("speak_session_audio", exist_ok=True)
VOICE_POOL = [
    {"name": "Rachel", "id": "21m00Tcm4TlvDq8ikWAM", "type": "professional female"},
    {"name": "Drew", "id": "29vD33N1CtxCmqQRPOHJ", "type": "confident male"},
    {"name": "Clyde", "id": "2EiwWnXFnvU5JabPnv8n", "type": "warm authoritative"},
    {"name": "Domi", "id": "AZnzlk1XvdvUeBnXmlld", "type": "energetic female"},
    {"name": "Dave", "id": "CYw3kZ02Hs0563khs1Fj", "type": "casual conversational"},
]

# 1. Retrieve Mavera personas
resp = requests.get(f"{MV_BASE}/personas", headers=MV_H).json()
personas = (resp if isinstance(resp, list) else resp.get("data", []))[:5]
print(f"Personas: {len(personas)}")

# 2. Map personas to voices
mappings = []
for i, p in enumerate(personas):
    v = VOICE_POOL[i % len(VOICE_POOL)]
    mappings.append({"persona_id": p["id"], "name": p.get("name", f"Persona {i+1}"),
                     "voice_id": v["id"], "voice_name": v["name"]})
    print(f"  {p.get('name', 'N/A'):30s} → {v['name']} ({v['type']})")

TOPIC = "What makes a marketing campaign truly memorable in 2026?"
tracks = []

for m in mappings:
    # 3. Generate in-character content
    chat = requests.post(f"{MV_BASE}/mave/chat", headers=MV_H, json={
        "message": f"You are {m['name']}. Respond to this discussion topic in 3-4 sentences, "
            f"staying in character. Speak naturally as in a roundtable.\n\nTopic: {TOPIC}",
        "persona_id": m["persona_id"],
    }).json()
    content = chat.get("content", "")
    print(f"\n[{m['name']}]: {content[:120]}...")
    time.sleep(1)

    # 4. Convert to audio with matched voice
    tts = requests.post(f"{EL_BASE}/text-to-speech/{m['voice_id']}",
        headers={"xi-api-key": EL_KEY, "Content-Type": "application/json"},
        json={"text": content, "model_id": "eleven_multilingual_v2",
              "voice_settings": {"stability": 0.45, "similarity_boost": 0.75,
                                 "style": 0.4, "use_speaker_boost": True}})
    if tts.status_code == 200:
        safe = m["name"].lower().replace(" ", "-").replace(",", "")[:30]
        path = f"speak_session_audio/{safe}.mp3"
        with open(path, "wb") as f: f.write(tts.content)
        tracks.append({"persona": m["name"], "voice": m["voice_name"],
                       "size_kb": len(tts.content) // 1024, "path": path})
        print(f"  → {path} ({len(tts.content) // 1024} KB)")
    time.sleep(2)

print(f"\nTotal: {len(tracks)} tracks, {sum(t['size_kb'] for t in tracks)} KB")

Example Output

  VP of Marketing   → Rachel  → vp-of-marketing.mp3 (67 KB)
  Product Manager   → Drew    → product-manager.mp3 (72 KB)
  Early-Stage Founder → Clyde → early-stage-founder.mp3 (58 KB)
  Gen Z Consumer    → Domi    → gen-z-consumer.mp3 (63 KB)

[VP of Marketing]: "Memorability comes from emotional resonance backed by data..."
[Gen Z Consumer]: "If it doesn't feel authentic, I scroll past in 0.3 seconds..."

Total: 4 tracks, 260 KB

Error Handling

Voice ID mismatch

Pre-made IDs are stable but can change. Verify with GET /voices before starting. If a mapped voice is missing, fall back to the next in the pool.

More personas than voices

Voices repeat via modulo if personas exceed the pool. Expand by fetching all voices from GET /voices and selecting based on gender/accent metadata.

Merging tracks

Concatenate with ffmpeg: ffmpeg -i "concat:track1.mp3|silence.mp3|track2.mp3" -c copy session.mp3. Generate gaps: ffmpeg -f lavfi -i anullsrc=r=44100:cl=mono -t 1 silence.mp3.