Proto-Dravidian is the linguistic reconstruction of the common ancestor of the Dravidian languages. It is thought to have differentiated into Proto-North Dravidian, Proto-Central Dravidian, and Proto-South Dravidian, although the date of diversification is still debated.
As a proto-language, the Proto-Dravidian language is not itself attested in the historical record. Its modern conception is based solely on reconstruction. It is suggested that the language was spoken in the 4th millennium BCE, and started disintegrating into various branches around 3rd millennium BCE.
Vowels: Proto-Dravidian contrasted between five short and long vowels: *a, *ā, *i, *ī, *u, *ū, *e, *ē, *o, *ō. The sequences *ai and *au are treated as *ay and *av (or *aw)
Consonants: Proto-Dravidian is reconstructible with the following consonant phonemes (Subrahmanyam 1983:p40, Zvelebil 1990, Krishnamurthi 2003):
The alveolar stop *ṯ in many daughter languages developed into an alveolar trill /r/. The stop sound is retained in Kota and Toda (Subrahmanyam 1983). Malayalam still retains the original (alveolar) stop sound in gemination. (ibid). In Old Tamil it took the enunciative vowel like the other stops. In other words, *ṯ (or *ṟ) did not occur word-finally without the enunciative vowel (ibid).
Velar nasal *ṅ occurred only before *k in Proto-Dravidian (as in many of its daughter languages). Therefore, it is not considered a separate phoneme in Proto-Dravidian. However, it attained phonemic status in languages like Malayalam, Gondi, Konda and Pengo due to the simplification of the original sequence *ṅk to *ṅ. (Subrahmanyam 1983)
The glottal fricative *h has been proposed by Bh. Krishnamurthi to account for the Old Tamil Aytam (Āytam) and other Dravidian comparative phonological phenomena (Krishnamurthi 2003).
The Northern Dravidian languages Kurukh, Malto and Brahui are not easily derivable from the traditional Proto-Dravidian phonological system. McAlpin (2003) proposes that they have branched off from an earlier stage of Proto-Dravidian than the conventional reconstruction, which would only apply to the other languages. He suggests reconstructing a richer system of dorsal stop consonants: