### Text-to-Speech Inference with TurkicTTS

Source: https://github.com/is2ai/turkictts/blob/master/README.md

This Python script demonstrates how to perform text-to-speech inference using a pretrained vocoder and acoustic model. Ensure the vocoder and acoustic model checkpoints are unzipped in the same directory. Specify the correct paths for the vocoder checkpoint, config file, and model path. The script supports multiple Turkic languages for synthesis.

```python
from parallel_wavegan.utils import load_model
from espnet2.bin.tts_inference import Text2Speech
from scipy.io.wavfile import write
from utils import normalization
import torch

fs = 22050
vocoder_checkpoint="parallelwavegan_male2_checkpoint/checkpoint-400000steps.pkl" ### specify vocoder path
vocoder = load_model(vocoder_checkpoint).to("cuda").eval()
vocoder.remove_weight_norm()

### specify path to the main model(transformer/tacotron2/fastspeech) and its config file
config_file = "exp/tts_train_raw_char/config.yaml"
model_path = "exp/tts_train_raw_char/train.loss.ave_5best.pth"

text2speech = Text2Speech(
    config_file,
    model_path,
    device="cuda", ## if cuda not available use cpu
    ### only for Tacotron 2
    threshold=0.5,
    minlenratio=0.0,
    maxlenratio=10.0,
    use_att_constraint=True,
    backward_window=1,
    forward_window=3,
    ### only for FastSpeech & FastSpeech2
    speed_control_alpha=1.0,
)
text2speech.spc2wav = None  ### disable griffin-lim

text = "merhaba"
### available options are azerbaijani, bashkir, kazakh, kyrgyz, sakha, tatar, turkish, turkmen, uyghur, uzbek
lang = "turkish"

text = normalization(text, lang)
with torch.no_grad():
    c_mel = text2speech(text)['feat_gen']
    wav = vocoder.inference(c_mel)
write("result.wav", fs, wav.view(-1).cpu().numpy())
```

--------------------------------

### normalization(x, lang="kazakh")

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts text in any of the ten supported Turkic languages into the Kazakh Cyrillic alphabet via an IPA intermediate representation. This is the primary entry point used before feeding text into the acoustic model. The `lang` parameter selects the source language; the default is "kazakh".

```APIDOC
## `normalization(x, lang="kazakh")` — Transliterate source-language text to Kazakh script for TTS

### Description
Converts text in any of the ten supported Turkic languages into the Kazakh Cyrillic alphabet via an IPA intermediate representation. This is the primary entry point used before feeding text into the acoustic model. The `lang` parameter selects the source language; the default is `"kazakh"`. Supported values: `"kazakh"`, `"turkish"`, `"azerbaijani"`, `"kyrgyz"`, `"uzbek"`, `"turkmen"`, `"tatar"`, `"bashkir"`, `"sakha"`, `"uyghur"`.

### Method
```python
from utils import normalization

# Turkish → IPA → Kazakh Cyrillic
text_tr = "merhaba"
kz_text = normalization(text_tr, lang="turkish")
print(kz_text)  # Output: Kazakh Cyrillic transcription ready for TTS model input

# Uzbek (Latin script)
text_uz = "salom dunyo"
kz_text_uz = normalization(text_uz, lang="uzbek")
print(kz_text_uz)

# Azerbaijani
text_az = "salam"
kz_text_az = normalization(text_az, lang="azerbaijani")
print(kz_text_az)

# Default: Kazakh — text passes through unchanged in script, rules applied
text_kz = "Қазақстан"
kz_text_kz = normalization(text_kz, lang="kazakh")
print(kz_text_kz)
```
```

--------------------------------

### kazakh_to_ipa(text) / ipa_to_kazakh(text)

Source: https://context7.com/is2ai/turkictts/llms.txt

Bidirectional conversion between Kazakh Cyrillic and IPA. `kazakh_to_ipa` maps every Kazakh letter to an IPA symbol and applies four phonological rules. `ipa_to_kazakh` reverses the mapping with corresponding anti-rules. These functions also serve as the final stage for all other language pipelines.

```APIDOC
## `kazakh_to_ipa(text)` / `ipa_to_kazakh(text)` — Kazakh Cyrillic ↔ IPA conversion

### Description
Bidirectional conversion between Kazakh Cyrillic and IPA. `kazakh_to_ipa` maps every Kazakh letter to an IPA symbol and applies four phonological rules: soft-[l] palatalization ([ł] rule), word-initial [w] insertion before [o]/[ö], [u]→[w] before consonants, and [ǐ]→[i] before consonants. `ipa_to_kazakh` reverses the mapping with corresponding anti-rules. These functions also serve as the final stage for all other language pipelines (via `ipa_to_kazakh`), since Kazakh is the TTS model's input script.

### Method
```python
import ipa_convert

# Forward: Kazakh → IPA
ipa = ipa_convert.kazakh_to_ipa("Қазақстан")
print(ipa)  # e.g., "qɑzɑqstɑn" (IPA symbols)

# Backward: IPA → Kazakh
kz = ipa_convert.ipa_to_kazakh(ipa)
print(kz)  # "қазақстан"

# Round-trip test (should print "input text and output text -- identical")
ipa_convert.test_kazakh("Қазақ тілі әдемі")
```
```

--------------------------------

### Convert Uyghur Latin to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Use `uyghur_to_ipa` to convert Uyghur Latin script to IPA and `ipa_to_uyghur` to convert IPA back to Uyghur Latin. Handles digraphs and special characters like apostrophes for glottal stops.

```python
import ipa_convert

# Latin ULY input
ipa = ipa_convert.uyghur_to_ipa("Uyghur xelqi")
print(ipa)

back = ipa_convert.ipa_to_uyghur(ipa)
print(back)  # "uyghur xelqi"

# Digraph and glottal stop
ipa2 = ipa_convert.uyghur_to_ipa("qoshni")
print(ipa2)

ipa_convert.test_uyghur("Uyghur tili ghayiti bay til")
```

--------------------------------

### Azerbaijani Latin to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Azerbaijani Latin-script text to IPA and back. Maps specific characters like [Cc]→[Ǯ], [Çç]→[ʆ], [Gg]→[ḡ], and [Əə]→[æ].

```python
import ipa_convert

ipa = ipa_convert.azerbaijani_to_ipa("Azərbaycan")
print(ipa)  # IPA with æ for ə

back = ipa_convert.ipa_to_azerbaijani(ipa)
print(back)  # "azərbaycan"

# Round-trip
ipa_convert.test_azerbaijani("Bakı gözəl şəhərdir")
```

--------------------------------

### Turkmen Latin to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Turkmen Latin-script text to IPA and back. Handles specific graphemes like [Çç]→[ʆ], [Jj]→[Ǯ], [Žž]→[ʒ], [Ňň]→[ŋ], [Şş]→[ʃ], [Ww]→[v], and [Ýý]→[j].

```python
import ipa_convert

ipa = ipa_convert.turkmen_to_ipa("Türkmenistan")
print(ipa)

back = ipa_convert.ipa_to_turkmen(ipa)
print(back)  # "türkmenistan"

# Test with Ý and Ň
ipa2 = ipa_convert.turkmen_to_ipa("ýurt")
print(ipa2)

ipa_convert.test_turkmen("Aşgabat gözel şäher")
```

--------------------------------

### Uzbek Latin to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Uzbek Latin-script text to IPA and vice-versa. Handles specific graphemes like 'ch', 'sh', 'ng', 'g'', and 'oʻ'.

```python
import ipa_convert

ipa = ipa_convert.uzbek_to_ipa("salom dunyo")
print(ipa)  # IPA output

back = ipa_convert.ipa_to_uzbek(ipa)
print(back)  # "salom dunyo"

# Grapheme cluster handling
ipa2 = ipa_convert.uzbek_to_ipa("o'zbek tili")
print(ipa2)

# Round-trip
ipa_convert.test_uzbek("O'zbekiston go'zal")
```

--------------------------------

### kyrgyz_to_ipa(text) / ipa_to_kyrgyz(text)

Source: https://context7.com/is2ai/turkictts/llms.txt

Handles the Kyrgyz Cyrillic alphabet, including the [Жж]→[Ǯ] digraph mapping. Applies the same four base rules as Kazakh plus two Kyrgyz-specific rules for velar/uvular alternation driven by vowel harmony.

```APIDOC
## `kyrgyz_to_ipa(text)` / `ipa_to_kyrgyz(text)` — Kyrgyz Cyrillic ↔ IPA conversion

### Description
Handles the Kyrgyz Cyrillic alphabet, including the [Жж]→[Ǯ] digraph mapping. Applies the same four base rules as Kazakh plus two Kyrgyz-specific rules: back-vowel context triggers k→q and g→ɣ substitutions (velar/uvular alternation driven by vowel harmony). This correctly models the phonological distinction between front-vowel /k,g/ and back-vowel /q,ɣ/ environments.

### Method
```python
import ipa_convert

ipa = ipa_convert.kyrgyz_to_ipa("Кыргызстан")
print(ipa)  # IPA with uvular stops in back-vowel contexts

back = ipa_convert.ipa_to_kyrgyz(ipa)
print(back)  # "кыргызстан"
```
```

--------------------------------

### turkish_to_ipa(text) / ipa_to_turkish(text)

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Turkish Latin-script text to IPA and back. Handles Turkish-specific characters and applies rules for soft-[l] (ł), [u]→[w] before consonants, and [i]→[i] disambiguation.

```APIDOC
## `turkish_to_ipa(text)` / `ipa_to_turkish(text)` — Turkish ↔ IPA conversion

### Description
Converts Turkish Latin-script text to IPA and back. Handles Turkish-specific characters (Ç/ç, Ş/ş, Ğ/ğ, İ/i, Ö/ö, Ü/ü) and applies rules for soft-[l] (ł), [u]→[w] before consonants, and [i]→[i] disambiguation. The [Cc] digraph is mapped to the affricate [Ǯ] (representing /dʒ/), and [Ğğ] to the fricative [ɣ]. Also experimentally supports circumflex vowels î and â.

### Method
```python
import ipa_convert

ipa = ipa_convert.turkish_to_ipa("merhaba")
print(ipa)  # "merhɑbɑ"

ipa2 = ipa_convert.turkish_to_ipa("Türk dünyası")
print(ipa2)

back = ipa_convert.ipa_to_turkish(ipa2)
print(back)  # "türk dünyası"

# Test round-trip
ipa_convert.test_turkish("Türkiye güzel bir ülkedir")
# Expected: "input text and output text -- identical"
```
```

--------------------------------

### Full TTS Inference Pipeline for Turkic Languages

Source: https://context7.com/is2ai/turkictts/llms.txt

Synthesize speech for any supported Turkic language using a pipeline that combines text normalization, an acoustic model (Tacotron2), and a neural vocoder (ParallelWaveGAN). Ensure vocoder and acoustic model checkpoints are downloaded and unzipped.

```python
from parallel_wavegan.utils import load_model
from espnet2.bin.tts_inference import Text2Speech
from scipy.io.wavfile import write
from utils import normalization
import torch

fs = 22050

# Load vocoder
vocoder_checkpoint = "parallelwavegan_male2_checkpoint/checkpoint-400000steps.pkl"
vocoder = load_model(vocoder_checkpoint).to("cuda").eval()
vocoder.remove_weight_norm()

# Load acoustic model (Tacotron2)
config_file = "exp/tts_train_raw_char/config.yaml"
model_path  = "exp/tts_train_raw_char/train.loss.ave_5best.pth"

text2speech = Text2Speech(
    config_file,
    model_path,
    device="cuda",            # use "cpu" if CUDA unavailable
    threshold=0.5,            # Tacotron2 stop threshold
    minlenratio=0.0,
    maxlenratio=10.0,
    use_att_constraint=True,
    backward_window=1,
    forward_window=3,
    speed_control_alpha=1.0,  # FastSpeech speed factor (ignored for Tacotron2)
)
text2speech.spc2wav = None   # disable Griffin-Lim; use neural vocoder instead

# --- Synthesize speech for each supported language ---
examples = [
    ("merhaba",         "turkish"),
    ("salom",           "uzbek"),
    ("Azərbaycan",      "azerbaijani"),
    ("Кыргызстан",      "kyrgyz"),
    ("Башҡортостан",    "bashkir"),
    ("Татарстан",       "tatar"),
    ("Türkmenistan",    "turkmen"),
    ("Саха",            "sakha"),
    ("Uyghur xelqi",    "uyghur"),
    ("Қазақстан",       "kazakh"),
]

for text, lang in examples:
    normalized = normalization(text, lang)          # transliterate to Kazakh script
    with torch.no_grad():
        c_mel = text2speech(normalized)("feat_gen") # mel spectrogram
        wav   = vocoder.inference(c_mel)            # waveform
    out_path = f"result_{lang}.wav"
    write(out_path, fs, wav.view(-1).cpu().numpy())
    print(f"[{lang}] '{text}' → saved to {out_path}")
```

--------------------------------

### Bashkir Cyrillic to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Bashkir Cyrillic text to IPA and back. Handles Bashkir-specific letters like [Ғғ]→[ɣ], [Ҡҡ]→[q], [Ҙҙ]→[z], and [Ҫҫ]→[s].

```python
import ipa_convert

ipa = ipa_convert.bashkir_to_ipa("Башҡортостан")
print(ipa)  # Note: Ҡ → q

back = ipa_convert.ipa_to_bashkir(ipa)
print(back)  # "башҡортостан"

# Ғ and Ҙ handling
ipa2 = ipa_convert.bashkir_to_ipa("Ғәжәп")
print(ipa2)

ipa_convert.test_bashkir("Башҡорт теле матур тел")
```

--------------------------------

### Tatar Cyrillic to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Tatar Cyrillic text to IPA and back. Includes Tatar-specific letters like [Җҗ]→[Ǯ] and handles vowels [Әә]→[æ], [Өө]→[ɵ], [Үү]→[ʏ].

```python
import ipa_convert

ipa = ipa_convert.tatar_to_ipa("Татарстан")
print(ipa)

back = ipa_convert.ipa_to_tatar(ipa)
print(back)  # "татарстан"

# Tatar-specific Җ
ipa2 = ipa_convert.tatar_to_ipa("Җир")
print(ipa2)  # Contains Ǯ

ipa_convert.test_tatar("Татар теле матур тел")
```

--------------------------------

### Full TTS Inference Pipeline

Source: https://context7.com/is2ai/turkictts/llms.txt

End-to-end speech synthesis for any Turkic language using a pipeline that combines text normalization, an acoustic model (ESPnet2 Tacotron2), and a vocoder (ParallelWaveGAN).

```APIDOC
## Full TTS Inference Pipeline

### Description
Combines `normalization()` from `utils.py` with ESPnet2 Tacotron2 and ParallelWaveGAN to synthesize speech. Download the pretrained vocoder (`parallelwavegan_male2_checkpoint`) and acoustic model (`kaztts_male2_tacotron2_train.loss.ave`) from the ISSAI website, unzip both into the same directory, then run the following.

### Usage
```python
from parallel_wavegan.utils import load_model
from espnet2.bin.tts_inference import Text2Speech
from scipy.io.wavfile import write
from utils import normalization
import torch

fs = 22050

# Load vocoder
vocoder_checkpoint = "parallelwavegan_male2_checkpoint/checkpoint-400000steps.pkl"
vocoder = load_model(vocoder_checkpoint).to("cuda").eval()
vocoder.remove_weight_norm()

# Load acoustic model (Tacotron2)
config_file = "exp/tts_train_raw_char/config.yaml"
model_path  = "exp/tts_train_raw_char/train.loss.ave_5best.pth"

text2speech = Text2Speech(
    config_file,
    model_path,
    device="cuda",            # use "cpu" if CUDA unavailable
    threshold=0.5,            # Tacotron2 stop threshold
    minlenratio=0.0,
    maxlenratio=10.0,
    use_att_constraint=True,
    backward_window=1,
    forward_window=3,
    speed_control_alpha=1.0,  # FastSpeech speed factor (ignored for Tacotron2)
)
text2speech.spc2wav = None   # disable Griffin-Lim; use neural vocoder instead

# --- Synthesize speech for each supported language ---
examples = [
    ("merhaba",         "turkish"),
    ("salom",           "uzbek"),
    ("Azərbaycan",      "azerbaijani"),
    ("Кыргызстан",      "kyrgyz"),
    ("Башҡортостан",    "bashkir"),
    ("Татарстан",       "tatar"),
    ("Türkmenistan",    "turkmen"),
    ("Саха",            "sakha"),
    ("Uyghur xelqi",    "uyghur"),
    ("Қазақстан",       "kazakh"),
]

for text, lang in examples:
    normalized = normalization(text, lang)          # transliterate to Kazakh script
    with torch.no_grad():
        c_mel = text2speech(normalized)("feat_gen") # mel spectrogram
        wav   = vocoder.inference(c_mel)            # waveform
    out_path = f"result_{lang}.wav"
    write(out_path, fs, wav.view(-1).cpu().numpy())
    print(f"[{lang}] '{text}' → saved to {out_path}")
```
```

--------------------------------

### Kazakh Cyrillic ↔ IPA Conversion

Source: https://context7.com/is2ai/turkictts/llms.txt

Bidirectional conversion between Kazakh Cyrillic and IPA, including phonological rules for palatalization and vowel/consonant alternations. Serves as the final stage for other language pipelines.

```python
import ipa_convert

# Forward: Kazakh → IPA
ipa = ipa_convert.kazakh_to_ipa("Қазақстан")
print(ipa)  # e.g., "qɑzɑqstɑn" (IPA symbols)

# Backward: IPA → Kazakh
kz = ipa_convert.ipa_to_kazakh(ipa)
print(kz)  # "қазақстан"

# Round-trip test (should print "input text and output text -- identical")
ipa_convert.test_kazakh("Қазақ тілі әдемі")
```

--------------------------------

### Kyrgyz Cyrillic ↔ IPA Conversion

Source: https://context7.com/is2ai/turkictts/llms.txt

Handles Kyrgyz Cyrillic, including the [Жж]→[Ǯ] mapping. Applies base rules plus Kyrgyz-specific rules for velar/uvular alternations driven by vowel harmony.

```python
import ipa_convert

ipa = ipa_convert.kyrgyz_to_ipa("Кыргызстан")
print(ipa)  # IPA with uvular stops in back-vowel contexts

back = ipa_convert.ipa_to_kyrgyz(ipa)
print(back)  # "кыргызстан"
```

--------------------------------

### Sakha (Yakut) Cyrillic to IPA and back

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Sakha Cyrillic text to IPA and back. Handles Sakha-specific features like [Ҕҕ]→[ɣ], [Ҥҥ]→[ŋ], and digraphs `дь`/`ДЬ`→[Ǯ], `нь`/`НЬ`→[ɲ].

```python
import ipa_convert

ipa = ipa_convert.sakha_to_ipa("Саха Өрөспүүбүлүкэтэ")
print(ipa)

back = ipa_convert.ipa_to_sakha(ipa)
print(back)  # "саха өрөспүүбүлүкэтэ"

# дь → Ǯ and нь → ɲ
ipa2 = ipa_convert.sakha_to_ipa("дьиэ")    # house
print(ipa2)  # Ǯɪě

ipa_convert.test_sakha("Саха тыла дэгэтэ")
```

--------------------------------

### Transliterate Turkic Text to Kazakh Script for TTS

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts text in supported Turkic languages to Kazakh Cyrillic via IPA. This is the primary entry point for TTS inference. The `lang` parameter specifies the source language.

```python
from utils import normalization

# Turkish → IPA → Kazakh Cyrillic
text_tr = "merhaba"
kz_text = normalization(text_tr, lang="turkish")
print(kz_text)  # Output: Kazakh Cyrillic transcription ready for TTS model input

# Uzbek (Latin script)
text_uz = "salom dunyo"
kz_text_uz = normalization(text_uz, lang="uzbek")
print(kz_text_uz)

# Azerbaijani
text_az = "salam"
kz_text_az = normalization(text_az, lang="azerbaijani")
print(kz_text_az)

# Default: Kazakh — text passes through unchanged in script, rules applied
text_kz = "Қазақстан"
kz_text_kz = normalization(text_kz, lang="kazakh")
print(kz_text_kz)
```

--------------------------------

### Uyghur Latin ↔ IPA Conversion

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Uyghur Latin-script text (ULY) to IPA and vice versa. Handles specific two-character graphemes and the apostrophe for glottal stops. The `ipa_to_uyghur` function restores `ë` before `e` to maintain accuracy.

```APIDOC
## `uyghur_to_ipa(text)` / `ipa_to_uyghur(text)`

### Description
Converts Uyghur Latin-script text (ULY — Uyghur Latin Yéziqi) to IPA and back. Handles two-character graphemes: `ch`→[ʆ], `sh`→[ʃ], `gh`→[ɣ], `ng`→[ŋ], `zh`→[ʒ], and `j`→[Ǯ]. The e-diaeresis `ë`/`é` maps to [e] (the more front vowel), while plain `e` maps to [æ]. Apostrophe `'` maps to glottal stop [ʔ]. In `ipa_to_uyghur`, [e] is carefully restored to `ë` before [æ]→`e` to handle precedence.

### Usage
```python
import ipa_convert

# Latin ULY input
ipa = ipa_convert.uyghur_to_ipa("Uyghur xelqi")
print(ipa)

back = ipa_convert.ipa_to_uyghur(ipa)
print(back)  # "uyghur xelqi"

# Digraph and glottal stop
ipa2 = ipa_convert.uyghur_to_ipa("qoshni")
print(ipa2)

ipa_convert.test_uyghur("Uyghur tili ghayiti bay til")
```
```

--------------------------------

### Turkish ↔ IPA Conversion

Source: https://context7.com/is2ai/turkictts/llms.txt

Converts Turkish Latin-script text to IPA and back, handling Turkish-specific characters and phonological rules. Supports digraphs like [Cc] and [Ğğ].

```python
import ipa_convert

ipa = ipa_convert.turkish_to_ipa("merhaba")
print(ipa)  # "merhɑbɑ"

ipa2 = ipa_convert.turkish_to_ipa("Türk dünyası")
print(ipa2)

back = ipa_convert.ipa_to_turkish(ipa2)
print(back)  # "türk dünyası"

# Test round-trip
ipa_convert.test_turkish("Türkiye güzel bir ülkedir")
# Expected: "input text and output text -- identical"
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.