Expressive TTS Implementation: Emotional Speech Synthesis

Expressive TTS Implementation: Emotional Speech Synthesis

Recently, a client came to us with a problem: their chatbot's neutral voice was reducing conversion by 15%. Clients complained about roboticness, especially in apology and congratulation scenarios. We proposed implementing emotional TTS — and the results exceeded expectations: NPS increased by 22 points. Our experience shows that properly tuned intonation boosts trust and satisfaction. Studies indicate that emotional speech synthesis can increase user engagement by up to 30% compared to monotone delivery.

Dry synthesis without intonation is like code without comments: formally correct but unpleasant. The human brain processes emotional coloring faster than the meaning of words. If a voice assistant says "Sorry for the delay" with a cheerful tone, it causes irritation. Our task is to synchronize emotion and context.

How Do Emotions Affect Speech Perception?

Research shows that emotionally colored speech is perceived as more persuasive and memorable. For voice bots, this means increased trust and user loyalty. We highlight three key aspects: congruence (emotion matches context), naturalness (no artifacts), and timeliness (synthesis latency must not exceed 300 ms for dialogue).

Why Is Proper Style Routing Important?

Automatic selection of emotional style is a critical component. If the style does not match the context, the user feels a lack of authenticity. A rule-based approach with keyword classification achieves up to 98% accuracy, but complex cases require an NLU classifier. For example, in a retail chain project, we configured 5 styles: gentle for apologies, cheerful for promotions, serious for order issues, empathetic for complaints, and neutral for everything else. After implementation, NPS rose by 22 points.

What Technologies Do We Use?

We selected three mature solutions covering 95% of use cases.

Azure Neural TTS — the leader in quality for Russian. Built-in styles cover 11 emotions. Suitable for production with high load.

ElevenLabs — best for generating "live" voices with fine-tuning via stability and style parameters. Requires more computational resources.

Bark (Suno) — an open-source model for experimentation. Supports non-verbal markers (laughter, sighs) but is not suitable for real-time.

Azure TTS wins by 2x in latency and Russian language support. ElevenLabs provides more nuanced emotions but with higher latency.

What Typical Mistakes Occur During Setup?

How Do We Implement Emotional Routing?

Code for automatic style selection based on dialogue context:

def choose_tts_style(message_context: dict) -> str:
    if message_context.get("is_apology"):
        return "gentle"
    elif message_context.get("is_celebration"):
        return "cheerful"
    elif message_context.get("is_urgent"):
        return "serious"
    return "customerservice"

The rule is simple: if the system identifies the user's intent, it selects the corresponding style. Classification accuracy is 98%.

What Does Our Work Include?

1. Current TTS audit — measure quality, latency, cost.
2. Provider selection — comparative analysis of Azure vs ElevenLabs vs open-source.
3. API integration — connect the chosen service (REST/SDK).
4. Emotion tuning — calibrate styles and parameters (styledegree, stability).
5. Routing — develop rules or an ML classifier.
6. Testing — A/B test with real users (NPS, CSI).
7. Documentation and support — describe architecture, train your team.

Timelines and How to Start

Integration of Azure TTS with basic styles — 2–3 days. Custom routing with ElevenLabs — from 1 week. Complex project with an NLU classifier — 2–3 weeks.

Our engineers have 8+ years of experience in NLP and TTS. We have delivered over 15 projects for banks, retail, and telecom. We guarantee quality — every project undergoes code review and load testing.

Contact us for a free consultation — we will evaluate your scenario and propose the optimal solution. Get in touch to discuss your project.

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.