Integration of SaluteSpeech TTS (Sber) for Speech Synthesis

A typical integration challenge with SaluteSpeech TTS is the short-lived access tokens (30 minutes) and Sber's non-standard SSL certificate chain. Without automatic token refresh, the service breaks every half hour; ignoring the certificate causes SSLError in production. We set up a full speech synthesis pipeline solving both. For one retailer, we deployed an on-premise solution with auto-refresh, reducing p99 latency from 800 ms to 450 ms due to local processing. The on-premise transition saved the company over 200,000 rubles annually in cloud request costs, and for large projects savings can reach 300,000 rubles per year.

Available SaluteSpeech TTS Voices

SaluteSpeech offers six built-in voices with different emotional tones. All voices synthesize speech at 24 kHz and support WAV16 format.

For long texts (more than 5000 characters), we automate chunking by sentence boundaries.

Why On-Premise Is Critical for Corporate Clients?

For companies with trade secret or state secret regimes, cloud connectivity is not allowed. SaluteSpeech TTS can be deployed inside the perimeter—data never leaves your network. We configure:

• Sber's root certificate for internal PKI
• Routing via internal load balancer
• Monitoring with Prometheus + Grafana (p99 latency, request count, auth errors)

On-premise also reduces dependency on external channels: cloud latency can reach 1.5 s, while local processing stays around 400–500 ms. For p99 latency, SaluteSpeech TTS on-premise is 1.6 times faster than cloud Yandex SpeechKit. Data from official SaluteSpeech and Yandex SpeechKit documentation.

How to Automate Token Refresh?

Here is a basic background worker in Python using apscheduler:

import requests
import base64
from apscheduler.schedulers.background import BackgroundScheduler

def refresh_token(client_id, client_secret):
    response = requests.post(
        "https://ngw.devices.sberbank.ru:9443/api/v2/oauth",
        headers={
            "Authorization": f"Basic {base64.b64encode(f'{client_id}:{client_secret}'.encode()).decode()}",
            "RqUID": "unique-uuid-here",
            "Content-Type": "application/x-www-form-urlencoded"
        },
        data={"scope": "SALUTE_SPEECH_CORP"},
        verify="/path/to/sber-root-ca.pem"
    )
    new_token = response.json()["access_token"]
    # Save to Vault or Redis with TTL
    return new_token

scheduler = BackgroundScheduler()
scheduler.add_job(refresh_token, 'interval', minutes=25, args=[CLIENT_ID, CLIENT_SECRET])
scheduler.start()

Refresh runs 5 minutes before token expiry, overwriting in a secure store. This guarantees uninterrupted synthesis uptime. On failure, the worker retries up to 3 times with exponential backoff.

Integration Process Step by Step

1. Scenario analysis – determine required voices, load, on-premise necessity.
2. Architecture design – choose REST API or gRPC, configure balancing and audio caching.
3. Microservice implementation – write token refresh worker, error handling, text chunking.
4. Integration testing – verify with your CRM, IVR, or chatbot.
5. Deploy and monitor – deploy in your environment, set up alerts for latency and errors.

What's Included in the Full Scope of Work

• Architecture design (REST API / gRPC, load balancing, audio caching)
• Synthesis microservice with auto token refresh and error handling
• SSL trust configuration for Sber API (install root certificate)
• Integration with your system (CRM, IVR, chatbot) via REST or queues
• Load testing (up to 100 RPS, p99 latency measurement)
• Operations documentation and monitoring scripts

Comparison of SaluteSpeech TTS and Yandex SpeechKit

The table shows: for corporate scenarios with security requirements, SaluteSpeech wins on latency and on-premise capability. However, for creative tasks Yandex offers more emotions. Get a consultation—we'll help you choose the right solution.

Estimated Timelines

From 2 to 5 days depending on complexity (on-premise, number of voices, integration with legacy systems). Pricing is custom—contact us for an estimate.

Our team has extensive market experience and 30+ projects with TTS systems (Yandex, Google, Amazon Polly, Sber). We guarantee correct handling of Sber's SSL specifics and stable synthesis uptime.

Contact us for a preliminary assessment—we'll send an integration example with your test phrase. Get a consultation on your scenario.

Learn more at the SaluteSpeech API official documentation.

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.