Custom Vocabulary Implementation for STT Systems

You integrate STT for a call center, but recognition of customer names and legal terms yields 40% errors. The model does not know "ОГРН", "СНИЛС", "ИНН". Every missed policy number or product code is lost data. Custom vocabulary is the fastest way to improve recognition of specific terms, names, and acronyms without retraining the model. It acts as a hint to the STT engine: "pay special attention to these words." Our certified engineers—over 8 years of experience and 30+ STT implementations—configure a vocabulary tailored to your domain in 2–4 hours. Word Error Rate reduction reaches 40% as early as the second day.

Why Custom Vocabulary Rather Than Model Retraining?

Retraining (fine-tuning) requires labeled audio data (at least 10 hours) and takes 2–4 weeks. Custom vocabulary delivers results in 1–2 days, reducing Word Error Rate by 30–50% for target terms. It does not change the architecture—you can modify the vocabulary on the fly without service downtime. For most business scenarios (order processing, legal consultations), this is sufficient.

Implementation for Major Providers

AWS Transcribe Custom Vocabulary

import boto3

transcribe = boto3.client('transcribe')

# Create vocabulary from file (S3)
transcribe.create_vocabulary(
    VocabularyName='corporate-terms-v1',
    LanguageCode='ru-RU',
    VocabularyFileUri='s3://my-bucket/vocabulary.txt'
)

# Format of vocabulary.txt:
# Phrase\tSoundsLike\tIPA\tDisplayAs
# Б-Ф-И-О\tbeh ef ee oh\t\tБФИО
# ИНН\tin en en\t\tИНН

Azure Custom Speech

# Add domain adaptation data via Azure Portal or REST API
# Supports: pronunciation dictionary, phrase list
import requests

phrase_list = {
    "kind": "PhraseList",
    "locale": "ru-RU",
    "phrases": ["ОГРН", "СНИЛС", "КПП", "расчётный счёт"]
}

Faster-whisper with Initial Prompt

model = WhisperModel("large-v3", device="cuda")

# Initial prompt helps the model focus on relevant vocabulary
initial_prompt = "ИНН, ОГРН, СНИЛС, КПП, расчётный счёт, генеральный директор."

segments, _ = model.transcribe(
    audio,
    initial_prompt=initial_prompt,
    language="ru"
)

The initial_prompt method works unreliably for long files—the prompt is processed only for the first window. For production, we recommend using the provider's built-in custom vocab.

Approach Comparison

Custom vocabulary works 10x faster than model retraining and provides sufficient accuracy for 90% of tasks.

How We Reduce WER by 40% in 2 Days

The process includes auditing the current STT, designing a domain vocabulary, implementing via the chosen provider's API, and A/B testing on 100+ audio files. We use sound similarity (SoundsLike) for acronyms and pronunciation variants (IPA) for complex words. The result is measurable accuracy gain without infrastructure changes. For one insurance project, WER on terms "ДМС", "ВЗР", "ОМС" dropped from 55% to 12%.

What If the Vocabulary Doesn't Help?

Sometimes the custom vocabulary yields less than 10% improvement—a signal of deeper issues: poor audio quality, model not adapted to noise, or context overwhelmed by homonymy. In such cases, we recommend combining vocabulary with lightweight fine-tuning or data augmentation. We perform diagnostics and propose the optimal strategy.

Setup Process

Keeping the Vocabulary Up to Date

• Versioning: every change is a new Git tag (v1.0, v1.1).
• Automatic updates: CI/CD ingests new terms from Jira/spreadsheet.
• Monitoring: alerts when accuracy drops by more than 5%.

A typical mistake is not accounting for homonyms. For example, "БФИО" may be recognized as "behfeeoh". In AWS Transcribe, the SoundsLike column is used for this.

Timeline: basic integration takes 1–2 days, including vocabulary population. We can evaluate your project within 2 days. Get a consultation on your case within 1 day—our engineers hold AWS AI and Azure AI Engineer certifications.

For detailed study, refer to the official documentation: AWS Transcribe Custom Vocabulary and Wikipedia: Speech recognition.

Speech Recognition and Synthesis: ASR, TTS, Voice Cloning

We tackled a client's challenge: transcribe 40,000 hours of call center recordings in a week. Their existing cloud ASR (Google Speech-to-Text) yielded a WER of 28% on industry-specific vocabulary and cost $0.006 per minute — prohibitively expensive at that volume. The goal was to reduce WER below 10% and switch to self-hosted inference. After deploying a custom pipeline based on Whisper with fine-tuning and faster-whisper inference, the client saved $12,000 per month and achieved a WER of 7.3%.

How does speech recognition ASR handle noisy call center recordings?

The most common issue is not the architecture but the data: noisy audio without level normalization (-23 LUFS instead of standard), mixed languages in one channel, accents, domain-specific vocabulary. Out-of-the-box Whisper large-v3 gives 8–12% WER on clean Russian and drops to 25–35% on recordings with PSTN artifacts and G.711 narrowband codec. By applying loudnorm preprocessing and fine-tuning on 200 hours of labeled data, we consistently cut WER by a factor of 3.

Typical problems we encounter

WER does not converge to the desired metric. Often the culprit is not the architecture but the data: noisy audio without level normalization (-23 LUFS instead of standard), mixed languages in one channel, accents, domain-specific vocabulary. Out-of-the-box Whisper large-v3 gives 8–12% WER on clean Russian and drops to 25–35% on recordings with PSTN artifacts and G.711 narrowband codec.

Diarization fails with more than two speakers. pyannote/speaker-diarization-3.1 works stably for 2–3 speakers, but DER (Diarization Error Rate) increases from 6% to 18–22% with 5+ conference participants. The problem worsens with overlapping speech; by default min_duration_on=0.1 cuts short interjections. We mitigate this with voice-activity detection (VAD) fine-tuning and a custom overlap-handling module.

Voice cloning — latency vs. quality. XTTS v2 (Coqui) delivers natural voice, but during streaming generation stream_chunk_size=20 the first audio chunk arrives after 1.4–2.0 seconds — unacceptable for interactive scenarios. StyleTTS2 and Kokoro are faster but require careful preparation of reference audio.

How do we solve it in practice?

The basic stack for a production pipeline:

• ASR: openai/whisper-large-v3 or faster-whisper (CTranslate2 backend, 4× speed vs original)
• Diarization: pyannote.audio 3.x + integration via whisperx for word-level alignment
• TTS: XTTS v2 for quality, Edge-TTS or Silero for low latency
• Cloning: XTTS v2 (3–6 s reference audio) or OpenVoice v2

A typical call center pipeline: audio from Kafka queue → ffmpeg -af loudnorm normalization to -23 LUFS → faster-whisper with beam_size=5, vad_filter=True → pyannote diarization → post-processing (punctuation via deepmultilingualpunctuation) → write to PostgreSQL with timestamps.

Case study from our practice. A fintech company with 12,000 calls per day. Initial WER on Russian with banking vocabulary — 22% (Google STT). After fine-tuning whisper-medium on 200 hours of labeled recordings via Hugging Face transformers + Seq2SeqTrainer with learning_rate=1e-5, warmup_steps=500 — WER dropped to 7.3%. Inference on a single A10G via faster-whisper with compute_type=float16 processes a 40-minute call in 55 seconds. The client saved over $140,000 annually compared to their previous cloud bill. Contact us for a free pilot estimate to see similar savings on your data.

How to fine-tune Whisper on domain data?

When a general model underperforms, fine-tuning is the first tool. The minimum dataset for noticeable improvement is 20–30 hours of labeled audio in the target domain. Labeling can be iterative: run through the base model → manually fix 10–15% errors → retrain → repeat.

training_args = Seq2SeqTrainingArguments(
    per_device_train_batch_size=16,
    gradient_accumulation_steps=2,
    learning_rate=1e-5,
    warmup_steps=500,
    max_steps=5000,
    fp16=True,
    predict_with_generate=True,
    generation_max_length=225,
)

Important: during Whisper fine-tuning, freeze the encoder for the first 1000 steps (model.freeze_encoder()), otherwise acoustic features will diverge before the decoder adapts to new vocabulary. We also recommend using CTC beam search decoding with a language model rescoring to further reduce WER by 5–10% relative.

Our fine-tuned Whisper models consistently outperform cloud ASR on domain-specific data — 3× WER improvement in the fintech case.

Speech synthesis: How to choose a model for your task?

For interactive bots requiring TTFB < 300 ms — Silero or Kokoro. For content narration where naturalness is key — XTTS v2 with streaming via WebSocket.

Our process and deliverables

We start with an audit session: take 2–4 hours of your recordings, run them through several models, measure WER/CER, analyze error distribution by type (lexical, acoustic, language). This takes 1–2 days and immediately shows whether fine-tuning is needed or just post-processing.

Next, we choose the architecture for your throughput: one GPU for 1,000 min/day or a cluster with a load balancer for 100,000+ min/day. Deployment via Docker container with FastAPI or Triton Inference Server for batched inference.

What you get after engagement:

• Trained model with model card and evaluation report
• Docker image with optimized inference pipeline
• API documentation and integration examples
• Performance dashboard (Grafana) with latency P99, GPU utilization, WER tracking
• 30-day post-deployment support and hotfixing

Timelines depend on complexity:

• Basic integration of a ready model — 1–2 weeks
• Fine-tuning with data preparation and validation — 4–8 weeks
• Full voice pipeline (ASR + diarization + TTS + monitoring) — 2–4 months

Project investments typically range from $20,000 to $80,000. Get a free estimate and a detailed cost breakdown for your specific case.

Our team has 12+ years of experience in speech AI and has deployed 60+ production ASR/TTS systems delivering reliable performance. Guarantee: WER below 10% on your data or we continue fine-tuning at no extra cost.

Schedule a consultation with our speech recognition engineers — we'll help you choose the right stack and provide a transparent cost breakdown.