Smart Call Analysis with AI

Problem: manual call processing is a bottleneck

Imagine you manage a 50-operator call center. Each day, 500 calls come in with an average duration of 4 minutes. To evaluate quality, you need to listen to at least 10% — that's 200 minutes daily. After each call, the manager fills out a report — another 5 minutes. Multiply that by 500 calls, and you get 41 hours per month just on reports. AI does all this in seconds. We have deployed such solutions for over ten projects, processing more than 500,000 calls and reducing analysis time by 70% compared to manual methods. Typical integration cost starts from $5,000, cutting manual analysis costs by 80%. You spend hours listening to recordings to identify issues in scripts or evaluate performance. Manager errors go unnoticed, and customers leave. We automate this process with AI integration.

What tasks do we solve?

• Transcription and sentiment analysis: after each call, we automatically get text, determine emotional tone (positive/negative/neutral), and highlight key topics.
• Smart routing: AI analyzes the history of inquiries and predicts which department can best handle the request — the call is redirected before connection.
• Search across recordings: the entire archive becomes searchable — need to find mentions of a competitor over the past month? A few queries in Elasticsearch.

How we do it: architecture and stack

Post-call analysis (quick start):

Completed call → PBX Webhook → Download recording → STT (Whisper or Vosk) → NLP (spaCy + transformers) → CRM

Real-time processing (via SIP trunk):

Call → PBX → SIP trunk → Media Gateway (RTP → PCM) → AI Server (VAD + ASR + NLU) → Action (routing/prompt)

For deployment we use:

• Models: OpenAI Whisper (large-v3), custom fine-tuned models based on BERT for classification.
• Infrastructure: vLLM for inference, ONNX Runtime for optimization, pgvector for embedding storage.
• MLOps: MLflow for experiment tracking, Kubeflow for pipelines.

Why post-call analysis is easier to deploy?

Post-call analysis does not require network changes — you only need to configure a Webhook in your PBX control panel. The real-time option is more complex: it requires a dedicated SIP trunk and a GPU server for real-time audio stream processing.

How AI improves call routing?

Real-time call analysis allows routing based on predictive analytics: customer sentiment, call history, intent. AI makes decisions in 400 ms (p99), imperceptible to speakers. For example, if a customer is irritated, the call is directed to the most experienced operator. This increases service level and reduces repeat calls. We implemented real-time scenarios in two projects — clients noted a 15% increase in satisfaction.

Process of work

1. Analytics: audit of current PBX, requirements gathering, selection of scenarios (post-call, real-time, or both).
2. Design: integration architecture, model selection, security setup.
3. Implementation: we write code in Python (FastAPI + Celery for async), integrate with CRM via REST/SOAP.
4. Testing: run on historical recordings, A/B test on 10% of calls. Achieve transcription accuracy of 95%+.
5. Deploy: containerization (Docker + Kubernetes), monitoring (Prometheus + Grafana).

What is included (deliverables)

• API documentation for integration schemas.
• Access to a test environment during development.
• Training your team on dashboards.
• Code warranty and SLA for error handling within 4 hours.

With over 10 PBX integration projects and 500k+ calls processed, we bring 5 years of experience in NLP and MLOps. Typical integration cost starts from $5,000, cutting manual analysis costs by 80%.

Timelines

• Post-call analysis: from 1 to 2 weeks.
• Real-time integration: from 1.5 to 2 months.
• Full cycle (both scenarios + dashboards): up to 3 months.

Integration cost is calculated individually after auditing your infrastructure.

Typical mistakes when doing it yourself

• Ignoring Webhook ACK: Mango Office retries the request — duplicate recordings. Solution: return HTTP 200 immediately.
• Incorrect request signature: hash mismatch blocks requests. Solution: use HMAC with key-sorted parameters.
• STT model overload: high latency and cost. Solution: enable VAD to avoid processing silence.

We have extensive experience in NLP and MLOps with over ten PBX integration projects. Order an audit of your PBX – it's free and takes 30 minutes. We will prepare a proposal with exact timelines and cost. Contact us for a consultation – we will assess your scenario and offer the optimal solution.

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.