kalpesh77's picture
Upload dataset
f4563ec verified
|
Raw
History Blame
8.96 kB
metadata
language: mr
dataset_info:
  - config_name: barakhadi
    features:
      - name: base
        dtype: string
      - name: vowel
        dtype: string
      - name: mark
        dtype: string
      - name: result
        dtype: string
    splits:
      - name: train
        num_bytes: 9300
        num_examples: 300
    download_size: 3524
    dataset_size: 9300
  - config_name: consonants
    features:
      - name: token_id
        dtype: int64
      - name: char
        dtype: string
      - name: ipa
        dtype: string
      - name: varg
        dtype: string
      - name: place
        dtype: string
      - name: manner
        dtype: string
      - name: voiced
        dtype: bool
      - name: aspirated
        dtype: bool
      - name: nasal
        dtype: bool
      - name: duration_ms
        dtype: int64
    splits:
      - name: train
        num_bytes: 2310
        num_examples: 37
    download_size: 4602
    dataset_size: 2310
  - config_name: default
    features:
      - name: token_id
        dtype: int64
      - name: char
        dtype: string
      - name: unicode
        dtype: string
      - name: ipa
        dtype: string
      - name: type
        dtype: string
      - name: duration_ms
        dtype: int64
      - name: f0_hz
        dtype: int64
      - name: f1_hz
        dtype: int64
      - name: f2_hz
        dtype: int64
      - name: f3_hz
        dtype: int64
      - name: example_word
        dtype: string
      - name: script
        dtype: string
    splits:
      - name: vowels
        num_examples: 13
      - name: consonants
        num_examples: 37
      - name: barakhadi
        num_examples: 300
      - name: sandhi_rules
        num_examples: 8
      - name: pitch_rules
        num_examples: 5
      - name: token_map
        num_examples: 50
      - name: duration_rules
        num_examples: 50
    download_size: 0.05
    dataset_size: 0.1
  - config_name: duration_rules
    features:
      - name: token_id
        dtype: int64
      - name: char
        dtype: string
      - name: base_duration_ms
        dtype: int64
      - name: context
        dtype: string
      - name: duration_ms
        dtype: int64
      - name: multiplier
        dtype: float64
      - name: rule
        dtype: string
    splits:
      - name: train
        num_bytes: 1964
        num_examples: 18
    download_size: 3790
    dataset_size: 1964
  - config_name: pitch_rules
    features:
      - name: context
        dtype: string
      - name: pattern
        dtype: string
      - name: hz_start
        dtype: int64
      - name: hz_peak
        dtype: int64
      - name: hz_end
        dtype: int64
      - name: hz_boost
        dtype: int64
      - name: description
        dtype: string
      - name: example
        dtype: string
    splits:
      - name: train
        num_bytes: 1070
        num_examples: 8
    download_size: 4216
    dataset_size: 1070
  - config_name: pitch_rules_v2
    features:
      - name: context
        dtype: string
      - name: pattern
        dtype: string
      - name: hz_start
        dtype: int64
      - name: hz_peak
        dtype: int64
      - name: hz_end
        dtype: int64
      - name: hz_boost
        dtype: int64
      - name: description
        dtype: string
      - name: example
        dtype: string
    splits:
      - name: train
        num_bytes: 1070
        num_examples: 8
    download_size: 4216
    dataset_size: 1070
  - config_name: sandhi_rules
    features:
      - name: rule_id
        dtype: string
      - name: type
        dtype: string
      - name: pattern
        dtype: string
      - name: input
        dtype: string
      - name: output
        dtype: string
    splits:
      - name: train
        num_bytes: 835
        num_examples: 8
    download_size: 2834
    dataset_size: 835
  - config_name: vowels
    features:
      - name: token_id
        dtype: int64
      - name: char
        dtype: string
      - name: ipa
        dtype: string
      - name: type
        dtype: string
      - name: duration_ms
        dtype: int64
      - name: f0
        dtype: int64
      - name: f1
        dtype: int64
      - name: f2
        dtype: int64
      - name: f3
        dtype: int64
    splits:
      - name: train
        num_bytes: 922
        num_examples: 13
    download_size: 4037
    dataset_size: 922
license: apache-2.0
tags:
  - marathi
  - phonology
  - tts
  - speech-synthesis
  - vedic-neural-geometry
  - devanagari
  - phonetics
  - formants
  - acoustic
  - linguistics
  - indic-languages
  - computational-phonology
size_categories:
  - n<1K
configs:
  - config_name: barakhadi
    data_files:
      - split: train
        path: barakhadi/train-*
  - config_name: consonants
    data_files:
      - split: train
        path: consonants/train-*
  - config_name: duration_rules
    data_files:
      - split: train
        path: duration_rules/train-*
  - config_name: pitch_rules
    data_files:
      - split: train
        path: pitch_rules/train-*
  - config_name: pitch_rules_v2
    data_files:
      - split: train
        path: pitch_rules_v2/train-*
  - config_name: sandhi_rules
    data_files:
      - split: train
        path: sandhi_rules/train-*
  - config_name: vowels
    data_files:
      - split: train
        path: vowels/train-*

मराठी व्याकरण आणि ध्वनी मॅट्रिक्स (Marathi Phonology Matrices)

🎯 उद्देश्य

हा डेटासेट मराठी भाषेच्या गणितीय ध्वनी संश्लेषणासाठी (Mathematical Speech Synthesis) तयार केला आहे. यामध्ये मराठी वर्णमालेचे संपूर्ण फोनोलॉजिकल विश्लेषण, अकौस्टिक प्रॉपर्टीज, आणि टोकनायझेशन मॅपिंग समाविष्ट आहे.

📊 संरचना (Structure)

Split विवरण ओळी
vowels स्वर + Formant Frequencies (F0-F3) 13
consonants व्यंजने + Articulation Features 37
barakhadi बाराखडी मॅट्रिक्स (C+V → CV) 300
sandhi_rules संधि नियम (Vowel + Consonant) 8
pitch_rules Pitch/Tone नियम 5
token_map Token ID → Character Mapping 50
duration_rules उच्चार काल (Acoustic Duration) 50

🔬 वैज्ञानिक पाया

Formant Frequencies

प्रत्येक स्वरासाठी F0 (मूल फ्रिक्वेन्सी), F1, F2, F3 (Formant) Hz मध्ये नोंदवले आहेत. हे ध्वनी संश्लेषण इंजिनमध्ये वापरता येतात.

Articulation Features

प्रत्येक व्यंजनासाठी — Place (स्थान), Manner (प्रयत्न), Voiced (नाद), Aspirated (महाप्राण), Nasal (अनुनासिक) — हे गुणधर्म Binary/Category स्वरूपात आहेत.

Token Scheme

Range वर्ग
0-12 स्वर (Vowels)
26-50 पंचम वर्ग व्यंजने
51-54 अंतःस्थ (Approximants)
55-58 ऊष्म (Fricatives)
59 विशेष — ळ
60-62 संयुक्त (Compound)

🚀 वापर (Usage)

from datasets import load_dataset

# Load dataset
ds = load_dataset("kalpesh77/marathi-phonology-matrices")

# स्वर माहिती
vowel = ds["vowels"][0]
print(vowel["char"])        # "अ"
print(vowel["f1_hz"])       # 750
print(vowel["duration_ms"]) # 120

# व्यंजन माहिती
consonant = ds["consonants"][0]
print(consonant["char"])     # "क"
print(consonant["place"])    # "velar"
print(consonant["voiced"])   # False

# बाराखडी शोधा
barakhadi = ds["barakhadi"].filter(lambda x: x["base_consonant"] == "क")
for row in barakhadi:
    print(f"{row['rule']}")

# संधि नियम
sandhi = ds["sandhi_rules"][0]
print(sandhi["example_output"])  # "देवेच्छा"

# Token ID → Character
import pandas as pd
token_df = pd.DataFrame(ds["token_map"])
print(token_df[token_df["token_id"] == 26])

🏗️ आराखडा (Pipeline)

टप्पा १: व्याकरण मॅट्रिक्स ✅ (हा डेटासेट)
टप्पा २: डिजिटल कोडमध्ये रूपांतर ✅ (Token Map)
टप्पा ३: उच्चार काल आणि ताण ✅ (Duration + Pitch)
टप्पा ४: गणितीय ध्वनी संश्लेषण इंजिन 🔄 (In Progress)
टप्पा ५: चाचणी आणि विस्तार 🔄 (In Progress)

🌐 विस्तार (Future Scope)

  • सामवेद (SamaVeda) — वैदिक स्वर आणि नाद
  • संस्कृत (Sanskrit) — शास्त्रीय उच्चार नियम
  • इतर भाषा — या मूळ संरचनेचा पुनर्वापर

📜 लायसन्स

Apache-2.0 — मुक्त वापर, शेअर, आणि सुधारणा.

👤 तयार केलेले

kalpesh77 — Vedic Neural Geometry Project


"मराठी ते गणितीय ध्वनी संश्लेषण — संपूर्ण आराखडा"