ElevenLabs API Case Study: How an Indie Game Studio Generated 200+ NPC Dialogue Lines in 48 Hours

From Casting Calls to API Calls: Replacing Traditional Voice Acting Pipelines

For indie game studios, voice acting is one of the most expensive and time-consuming production bottlenecks. Casting agencies, recording sessions, retakes, and post-processing can consume weeks of calendar time and thousands of dollars — even for a modest RPG with a handful of NPCs. This case study documents how a fictional but representative indie studio, Ironpine Games, used the ElevenLabs API to generate over 200 fully voiced NPC dialogue lines in just 48 hours. The workflow leveraged three core ElevenLabs features: Projects API, Voice Design presets, and Pronunciation Dictionaries — replacing what would have traditionally required a 3-week casting and recording pipeline.

The Challenge

• 212 dialogue lines across 14 unique NPCs for a fantasy RPG vertical slice- Budget constraint: under $500 total voice production cost- Timeline: 48 hours before a publisher demo- Each NPC needed a distinct, consistent voice with correct pronunciation of 30+ fictional proper nouns (place names, spells, lore terms)

Step 1: Environment Setup and Installation

Install the ElevenLabs Python SDK

pip install elevenlabs

Configure Your API Key

# Set your API key as an environment variable
export ELEVENLABS_API_KEY=YOUR_API_KEY

# Python initialization
from elevenlabs.client import ElevenLabs
client = ElevenLabs(api_key=“YOUR_API_KEY”)

Step 2: Design Unique NPC Voices with Voice Design

Instead of auditioning voice actors, Ironpine used the Voice Design API to generate distinct voice profiles for each NPC archetype — grizzled blacksmith, young apprentice, ancient oracle, and so on. from elevenlabs import VoiceDesign, Gender, Age, Accent

Design a grizzled blacksmith voice

print(f”Created voice: {blacksmith_voice.voice_id}“)

The team repeated this for all 14 NPCs, generating 2-3 preview variations per character and selecting the best fit — a process that took roughly 3 hours compared to weeks of casting calls.

Step 3: Create a Pronunciation Dictionary for Lore Terms

Fantasy games are full of invented words. Without a pronunciation dictionary, the TTS engine will guess — often incorrectly. ElevenLabs Pronunciation Dictionaries solve this definitively. import json

print(f”Dictionary ID: {dictionary.id}”) print(f”Rules added: {dictionary.version_id}“)

Example PLS Lexicon File

  
Valdrethar
vɑːl.drɛ.θɑːr
  
  
Kythira
kɪ.θaɪ.rə
  
  
Aethermancy
iː.θɜːr.mæn.si
  

Step 4: Batch Generate All Dialogue with the Projects API

The Projects API is where the entire pipeline comes together. It allows you to organize chapters, assign voices per character, attach pronunciation dictionaries, and batch-convert an entire script. # Create a project for the RPG vertical slice project = client.projects.add( name="Ironpine RPG - Vertical Slice", default_model_id="eleven_multilingual_v2", pronunciation_dictionary_versions_locators=[ {"pronunciation_dictionary_id": dictionary.id, "version_id": dictionary.version_id} ], default_paragraph_voice_id=blacksmith_voice.voice_id )

print(f”Project created: {project.project_id}”)

# Add a chapter for each game area or quest
chapter = client.projects.add_chapter(
project_id=project.project_id,
name=“Chapter 1 - Village of Valdrethar”
)

print(f”Chapter ID: {chapter.chapter_id}“)

Bulk Upload Dialogue Lines via Script

import csv
import time

npc_voices = {
    "Gorath": "voice_id_blacksmith",
    "Lyra": "voice_id_apprentice",
    "Elder Morvyn": "voice_id_oracle",
    # ... 11 more NPC voice mappings
}

with open("dialogue_script.csv", "r") as f:
    reader = csv.DictReader(f)  # columns: npc_name, line_id, text
    for row in reader:
        voice_id = npc_voices.get(row["npc_name"])
        if not voice_id:
            continue

        audio = client.text_to_speech.convert(
            voice_id=voice_id,
            text=row["text"],
            model_id="eleven_multilingual_v2",
            pronunciation_dictionary_locators=[
                {"pronunciation_dictionary_id": dictionary.id, "version_id": dictionary.version_id}
            ]
        )

        filename = f"audio/{row['line_id']}.mp3"
        with open(filename, "wb") as out:
            for chunk in audio:
                out.write(chunk)

        print(f"Generated: {filename}")
        time.sleep(0.5)  # respect rate limits

Results Summary

Can I use ElevenLabs-generated voices commercially in a shipped game?

Yes. ElevenLabs allows commercial usage of generated audio on paid plans. The voices created through Voice Design are fully owned synthetic voices with no likeness rights concerns, making them ideal for indie game distribution on Steam, itch.io, or console storefronts. Always review the current terms of service for your specific plan tier.

How do I maintain voice consistency when generating hundreds of lines over multiple sessions?

Once you save a designed voice via create_voice_from_preview, it receives a persistent voice_id. All subsequent TTS calls using that ID produce consistent output. Keep stability at 0.5 or higher and use the same model_id across all generations. Avoid regenerating the voice profile mid-production.

What happens if I need to add new dialogue lines after the initial batch?

Simply run the same script with additional CSV rows. The voice IDs, pronunciation dictionary, and model settings remain unchanged. New lines will sound consistent with previously generated audio. For large additions, consider using the Projects API to organize new content into separate chapters for easier management.