DSE-TTS: Dual Speaker Embedding for Cross-Lingual Text-to-Speech