Customers increasingly ignore mass SMS and email campaigns — open rates drop to 10–20%. A voice bot personalized with purchase history boosts conversion 3–5× compared to text channels. We develop AI auto-dialer systems that not only inform about promotions but also engage in conversation and place orders during the call. Over 5 years, we have completed 120+ deployments for retail, banking, and telecom.
Key Problems Solved by the System
Low conversion of mass campaigns. SMS and email open rates are 10–20%, while 70–80% answer a call. A voice bot holds attention up to 45 seconds, and a personalized offer based on purchase history drives orders.
Lack of personalization. We build the promo message dynamically: considering the customer's favorite category, average order value, and recency of last purchase. The bot uses a RAG approach RAG: it retrieves the relevant promotion from the database and injects it into a template.
Legal non-compliance. The system checks consent against the opt-in database, logs opt-outs, and respects time restrictions. Compliance is embedded at every stage.
How We Do It: Case Study of a Hypermarket Chain
A retail chain wanted to boost footfall for a "30% discount on home appliances" promotion. Previously, they sent SMS codes — conversion was 2.3%. We deployed a voice bot with dialogue capabilities.
Stack: Python + FastAPI, LLM GPT-4o for scenario generation, 1536-dim Embeddings for customer-specific promotion retrieval, Silero TTS, CRM integration via REST API.
The bot loads customer history, uses RAG to find a suitable promotion, synthesizes speech, and asks clarifying questions. If the customer agrees, the bot sends an SMS with a payment link or connects a live agent. Conversion rose to 11.4%, campaign ROI was 340%.
async def build_promo_message(customer: dict, promo: dict) -> str:
"""Build a personalized message"""
personalization = await get_personalization(customer)
if personalization.get("favorite_category"):
category_mention = f"for {personalization['favorite_category']}"
elif personalization.get("last_purchase"):
category_mention = "for items similar to your last order"
else:
category_mention = ""
template = f"""
Hello, {customer['first_name']}!
We have a special offer {category_mention}:
{promo['headline']}.
{promo['discount_description']}.
The offer is valid until {promo['expiry_date']}.
Would you like to learn more or place an order right now?
"""
return template.strip()
Why an AI Auto-Dialer Outperforms Text Campaigns?
Comparison of key metrics:
| Channel |
Average Conversion |
Contact Duration |
Dialogue Capability |
| SMS |
2–5% |
5 sec (reading) |
No |
| Email |
1–4% |
10 sec (scan) |
No |
| Voice Bot |
10–15% |
30–45 sec |
Yes |
The voice bot achieves 3–5× better conversion and also gathers additional customer insights during the conversation. Cost per lead can drop by up to 40% compared to SMS campaigns.
How to Comply with Advertising Call Regulations?
The system automatically checks the number against the consent database (Article 18 of the Federal Law “On Advertising”). If no consent, the call is blocked. Each call records the time: weekdays 9–20, weekends 10–17. At the start, the bot introduces itself and states the purpose; at the end, it offers an opt-out via the “Stop” command or SMS. All opt-outs are immediately added to a blocklist. Penalty savings from compliance can be significant.
Process and Timelines
| Stage |
Description |
Typical Duration |
| Analytics |
Study audience, purchase history, scenarios |
3–5 days |
| Design |
Define segments, templates, dialogue flows |
3–5 days |
| Development |
Configure LLM, TTS, CRM integration |
7–10 days |
| Testing |
A/B tests on part of the base, iterate |
4–5 days |
| Deployment |
Rollout, documentation, training |
2–3 days |
Basic version: 3–4 weeks. Extended (with A/B tests and ROI analytics): up to 1.5 months. The cost is calculated individually based on your integrations. Contact us for a free consultation.
Example Dialogue with a Customer
Bot: “Hello, Anna! This is Electronics-24. We have a promotion on smartphones — up to 30% off last year's models. Are you interested in smartphones?”
Customer: “Yes, which brand?”
Bot: “Samsung and Xiaomi. Which one interests you?”
Customer: “Samsung.”
Bot: “Great! We have the Galaxy A54 with 2000 cashback points. Would you like to order?”
Customer: “Yes.”
Bot: “We’ve sent a payment link via SMS. Thank you for your order!”
What's Included in the Deliverable
- A working voice bot with personalization and segmentation.
- Integration with your CRM (REST API, WebSocket, or message queue).
- Analytics dashboard showing conversion, ROI, and call logs.
- Operation and administration manual.
- 2-hour operator training.
- One month of post-launch support.
Request a pilot launch — we'll run an A/B test on 1,000 customers and demonstrate the real conversion lift.
Why Choose Our AI Auto-Dialer?
We guarantee transparent reporting, full compliance with 152-FZ, and up to 50% budget savings on marketing campaigns compared to traditional methods. Get a free project assessment — contact us today.
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.
| Model |
WER (clean) |
WER (noisy) |
RTF (A10G) |
Languages |
| Whisper large-v3 |
5.2% |
27% |
0.08 |
99 |
| Wav2Vec2-XLSR-53 |
6.8% |
32% |
0.12 |
143 |
| Google STT (cloud) |
7.0% |
28% |
– |
125 |
| DeepSpeech 0.9.3 |
11.5% |
41% |
0.06 |
8 |
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?
| Model |
Latency (TTFB) |
Naturalness MOS |
Cloning |
Languages |
| XTTS v2 |
1.2–2.0 s |
4.1–4.3 |
Yes, 3 s reference |
17 |
| StyleTTS2 |
0.3–0.6 s |
4.0–4.2 |
Yes, requires adaptation |
en, + fine-tune |
| Kokoro-82M |
0.08–0.15 s |
3.7–3.9 |
No |
en, ja |
| Silero TTS |
0.05–0.1 s |
3.4–3.6 |
No |
ru, en, de, etc. |
| Edge-TTS |
~0.4 s (cloud) |
4.0 |
No |
100+ |
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.
