Get Audio file via Command

[miguelgh65]

I use this command to get an audio file from cli but here the audio file Im getting is not saying anything at all, via GUI it works perfectly.
Im using a more extended alphabet than english
python synthesis/synthesize.py -m data/models/..... -vm "data/hifigan/vocoder/model.pt" -hc "data/hifigan/vocoder/config.json" -t "test this is a test" -a audio.wav
Any idea?
Thanks in advance

[scheissegalo]

I have the same issue and looking for an answer. I think the problem here is when using the synthesize.py there is no option to add the language model to. Default is english and no docs pointing on how to load the language model you made with the GUI.

synthesize.py:
from training import DEFAULT_ALPHABET ......... symbols=DEFAULT_ALPHABET

[scheissegalo]

Fixed by adding following code to the import section of synthesize.py:

from dataset.transcribe import Silero, DeepSpeech, SILERO_LANGUAGES
from main import app, paths
from training.utils import (
    get_available_memory,
    get_gpu_memory,
    get_batch_size,
    load_symbols,
    generate_timelapse_gif,
    create_trainlist_vallist_files,
)

TRANSCRIPTION_MODEL = "model.pbmm"
ALPHABET_FOLDER = "alphabets"
ALPHABET_FILE = "alphabet.txt"

def get_symbols(language):
    if language in SILERO_LANGUAGES:
        return load_symbols(os.path.join(ALPHABET_FOLDER, f"{language}.txt"))
    else:
        return load_symbols(os.path.join(paths["languages"], language, ALPHABET_FILE))

then you can add your language generated from gui into "def synthesize":

def synthesize(
    model,
    text,
    #symbols=DEFAULT_ALPHABET,
    symbols = get_symbols("<name of your language>"),
    graph_path=None,
    audio_path=None,
    vocoder=None,
    silence_padding=0.15,
    sample_rate=22050,
    max_decoder_steps=1000,
    split_text=False,
):