AI Audio Mastering: Turnkey Automation for Podcasts & Music

Problem: Manual mastering doesn’t scale

A podcast studio releases 50 episodes per week. Each track—interview recording, musical transition, ad insertion. Manual mastering takes 40 minutes per episode. Multiply by 50: 33 person-hours per week. That’s expensive and slow. For serial content, manual processing is unacceptable—automation is needed.

We build AI mastering pipelines that replace manual labor with conveyor-belt processing: loudness normalization, frequency correction, compression, and limiting. Our engineers hold audio processing certifications and have over 5 years of MLOps experience. With more than 50 audio processing projects completed, including automation for major podcast studios, AI mastering processes a track 30 times faster than manual, and cost per track drops by an order of magnitude at volumes above 500 tracks. Contact us for a free consultation on your project.

How Matchering adjusts a track to a reference

import matchering as mg

def master_to_reference(
    target_path: str,
    reference_path: str,
    output_path: str
) -> None:
    """Master target to sound like reference"""
    mg.process(
        target=mg.pcm16(target_path),
        reference=mg.pcm16(reference_path),
        results=[
            mg.Result(output_path, subtype="PCM_16"),
        ]
    )

Matchering analyzes the spectral and dynamic characteristics of the reference track and applies EQ + compression to the target so they sound similar. This method is especially useful for unifying the sound across tracks.

Why podcast mastering needs LUFS targeting

import subprocess
import json

def loudnorm_two_pass(input_path: str, output_path: str, target_lufs: float = -14.0) -> None:
    """
    -14 LUFS = Spotify/Apple Music
    -16 LUFS = YouTube
    -23 LUFS = EBU R128 (broadcast)
    """
    # Pass 1: analysis
    probe = subprocess.run([
        "ffmpeg", "-i", input_path,
        "-af", f"loudnorm=I={target_lufs}:TP=-1.5:LRA=11:print_format=json",
        "-f", "null", "-"
    ], capture_output=True, text=True)

    # Parse statistics from stderr
    stats = json.loads(probe.stderr.split("Parsed_loudnorm")[1].split("\n", 2)[2])

    # Pass 2: final normalization with measured parameters
    subprocess.run([
        "ffmpeg", "-i", input_path,
        "-af", (
            f"loudnorm=I={target_lufs}:TP=-1.5:LRA=11"
            f":measured_I={stats['input_i']}"
            f":measured_LRA={stats['input_lra']}"
            f":measured_TP={stats['input_tp']}"
            f":measured_thresh={stats['input_thresh']}"
            ":linear=true:print_format=summary"
        ),
        "-ar", "44100", output_path
    ], check=True)

EBU R128 is the loudness standard for broadcasting followed by most platforms. Failing LUFS compliance can result in track rejection. We use loudnorm with a two-pass scheme for precise compliance.

How automatic equalization works

import librosa
import numpy as np
from scipy.signal import butter, filtfilt

class AutoEqualizer:
    """Simple automatic EQ based on spectrum analysis"""

    TARGET_SPECTRUM = {
        "podcast": {
            100: -3,    # reduce rumble
            250: -2,    # clean muddiness
            3000: +2,   # vocal presence
            8000: +1,   # air
        },
        "music": {
            60: +2,
            200: -1,
            3000: +1,
            10000: +2,
        }
    }

    def analyze_and_correct(self, audio: np.ndarray, sr: int, profile: str = "podcast") -> np.ndarray:
        spectrum = np.abs(librosa.stft(audio))
        freqs = librosa.fft_frequencies(sr=sr)

        corrections = self.TARGET_SPECTRUM.get(profile, {})
        corrected = audio.copy()

        for freq_hz, gain_db in corrections.items():
            gain_linear = 10 ** (gain_db / 20)
            # Apply peaking filter around target frequency
            b, a = self._peak_filter(freq_hz, sr, gain_db, Q=2.0)
            corrected = filtfilt(b, a, corrected)

        return corrected

    def _peak_filter(self, freq, sr, gain_db, Q=2.0):
        # Implementation omitted for brevity
        pass

AutoEqualizer adjusts spectral balance to content type: for podcasts it reduces rumble and muddiness, for music it boosts lows and highs. Profiles can be extended for specific tasks.

What to choose: self-hosted or cloud AI mastering?

Self-hosted option pays for itself in 2-3 months when processing 500+ tracks per month. Learn more about matchering.

Comparison of manual and AI mastering

How to evaluate AI mastering quality?

For objective comparison we use metrics: LUFS (integrated loudness), True Peak (peak level), dynamic range (DR), and spectral centroid. We conduct A/B testing on a sample of 10-20 tracks: compare original, API mastering, and our pipeline. Based on results, we optimize compression, limiting, and EQ parameters. If needed, manual adjustments are applied—ensuring quality matches commercial services.

Our process for AI mastering

1. Analytics — audit current pipeline, gather requirements for loudness, formats, performance.
2. Design — select algorithms (matchering, loudnorm, AutoEQ), design pipeline architecture, prototype on samples.
3. Implementation — write code, integrate with your system (API, queue).
4. Testing — A/B comparison with manual mastering, measure metrics (LUFS, True Peak, latency p99).
5. Deployment — install on your servers or cloud, provide documentation, train your team.

Common mistakes in auto-tuning

• Using a single-pass loudnorm without measurement — loses accuracy.
• Wrong target LUFS for platform (e.g., -14 for YouTube).
• Ignoring True Peak — clipping after normalization.
• No presets for different music genres (pop requires more aggressive compression than classical).

Timeline and cost

Base pipeline (matchering + loudnorm) — from 2 weeks. With web interface and queue — 3-4 weeks. Cost is calculated individually based on integration complexity and required performance. We’ll evaluate your project for free—contact us for a consultation.

What’s included

• Source code of the pipeline (Python, Bash)
• Documentation for launch and configuration
• Docker image for deployment
• API documentation (OpenAPI)
• Training for your team (1-2 days)
• Support for 3 months after delivery

We guarantee stable pipeline operation under load up to 1000 tracks per day. We use only proven open-source libraries, eliminating vendor lock-in. Code undergoes review, is covered by tests, and is deployed in Docker. Order a pilot project—get a ready-made solution for your content.

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.