Skip to content

Turkish Dialect FT: Karadeniz / Aegean / Eastern Anatolian Pronunciation + Dataset Collection

Standard Turkish baseline Whisper is good but struggles with dialects (Black Sea 'cik' suffix, Eastern Anatolian, Aegean). Dialect speech recording protocol (consent), 50-100h regional corpus, FT + per-dialect WER. Production: customer service, healthcare (village services).

Şükrü Yusuf KAYA
26 min read
Advanced
Türkçe Lehçe FT: Karadeniz / Ege / Doğu Anadolu Telaffuzu + Dataset Toplama

1. Dataset Toplama Protokolü (Rıza + KVKK)#

1. Gönüllü konuşmacılar (region başına 30-50 kişi)
2. Yazılı rıza formu (KVKK uyumlu):
   - Ses kaydının amacı (ASR FT)
   - Veri saklama süresi
   - Anonimleştirme (PII strip)
   - Geri çekme hakkı (silme)
3. Kayıt protokolü:
   - 16 kHz mono, 30-90 dk per session
   - Read speech (script) + spontaneous (free talk)
   - Multiple acoustic conditions (sessiz oda, gürültülü)
4. Annotation:
   - Forced alignment (Montreal Forced Aligner TR)
   - Verbatim transcription (lehçe formu)
   - Standart TR'ye çeviri (opsiyonel translation FT için)
5. Dataset card + datasheet (HuggingFace format)
Maliyet: 50 saat regional corpus ~₺50K-100K (kayıt + annotation + legal).
✅ Teslim
  1. Bir region için 1 saat pilot corpus topla. 2) Whisper FT baseline vs region-FT WER karşılaştır. 3) Sonraki ders: 7.4 — Streaming ASR (faster-whisper, distil-whisper).

Yorumlar & Soru-Cevap

(0)
Yorum yazmak için giriş yap.
Yorumlar yükleniyor...

Related Content

Part 0 — Engineering Foundations

Welcome to the Fine-Tuning Cookbook: System, Stage Taxonomy, and the Reproducibility Contract

Start Learning
Part 0 — Engineering Foundations

Reproducibility Stack: Seeds, cuDNN Flags, and Deterministic CUDA — End the 'Works on My Machine' Problem

Start Learning
Part 0 — Engineering Foundations

Environment Pinning: uv + pyproject.toml, CUDA Version Matrix, and Container Recipes

Start Learning