[hts-users:03390] Re: measure of labeled data for initialization
- Subject: [hts-users:03390] Re: measure of labeled data for initialization
- From: Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
- Date: Thu, 9 Aug 2012 02:44:59 +0900
- Cc: uratec <uratec@xxxxxxxxxxxx>
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20120113; h=mime-version:sender:in-reply-to:references:date :x-google-sender-auth:message-id:subject:from:to:cc:content-type :content-transfer-encoding; bh=72Gb9en1f2MnQIKQFkOGQKkAh/QV0foGSDS2q0b8K78=; b=e+/Q4ptccxiFdxlTtZyBrXVYC/IJ2fhyiaISXBMoAMbLuC+ynmGPCwQtigxES4tJYu ZalU4aAEBR7B8pdm5BlpPAIEdoQfN1mcS7FcX9b4NVSR2zoKpJurupPeWNJJZOz0JOw/ KFEMC0jeIglPpTxE3koxSXLyfpC9QAO5KuUGbt0V7Pp8pQvZKW5+71mH7MQyw78m5gID tlWBitflUkTaNK8eAF3YNUv6k+93ok9S96CTd2sk8BuBKwhB3y36tvW6A0VLlV4UJb+9 +bvvZanaVcFItqxPXhf/0/madiFFUHLY8Z+4w3YGzSdNe4NNDlPdrWUW07W/QsEwOVGN +QHA==
Hi,
If you have correct labeled data, your technique may work well.
But, if your labeled data don't have correct segmentation, I recommend
the flat start technique with DAEM algorithm.
Regards,
Keiichiro Oura
2012/7/12 Ramil Məmmədov <ramil2319@xxxxxxxxx>:
> Dear HTS users,
> I want to devolop models with labeled and unlabeled data in my language. So,
> in initialization i used labeled data and after i used unlabeled data. Does
> anyone have any suggestions about measure of labeled data on initialization?
> For example, can i use 20 labeled and 100 unlabeled utterance for "auu"
> phonem?
- References
-
- [hts-users:03383] measure of labeled data for initialization, Ramil Məmmədov